LONDON MATHEMATICAL SOCIETY LECTURE NOTE SERIES

Managing Editor: Professor M. Reid, Mathematics Institute,

University of Warwick, Coventry CV4 7AL, United Kingdom

The titles below are available from booksellers, or from Cambridge University Press at http://www.cambridge.org/mathematics

337 Methods in Banach space theory, J.M.F. CASTILLO & W.B. JOHNSON (eds)
338 Surveys in geometry and number theory, N. YOUNG (ed)
339 Groups St Andrews 2005 I, C.M. CAMPBELL, M.R. QUICK, E.F. ROBERTSON & G.C. SMITH (eds)
340 Groups St Andrews 2005 II, C.M. CAMPBELL, M.R. QUICK, E.F. ROBERTSON & G.C. SMITH (eds)
341 Ranks of elliptic curves and random matrix theory, J.B. CONREY, D.W. FARMER, F. MEZZADRI &
N.C. SNAITH (eds)
342 Elliptic cohomology, H.R. MILLER & D.C. RAVENEL (eds)
343 Algebraic cycles and motives I, J. NAGEL & C. PETERS (eds)
344 Algebraic cycles and motives II, J. NAGEL & C. PETERS (eds)
345 Algebraic and analytic geometry, A. NEEMAN
346 Surveys in combinatorics 2007, A. HILTON & J. TALBOT (eds)
347 Surveys in contemporary mathematics, N. YOUNG & Y. CHOI (eds)
348 Transcendental dynamics and complex analysis, P.J. RIPPON & G.M. STALLARD (eds)
349 Model theory with applications to algebra and analysis I, Z. CHATZIDAKIS, D. MACPHERSON,
A. PILLAY & A. WILKIE (eds)
350 Model theory with applications to algebra and analysis II, Z. CHATZIDAKIS, D. MACPHERSON,
A. PILLAY & A. WILKIE (eds)
351 Finite von Neumann algebras and masas, A.M. SINCLAIR & R.R. SMITH
352 Number theory and polynomials, J. MCKEE & C. SMYTH (eds)
353 Trends in stochastic analysis, J. BLATH, P. MÖRTERS & M. SCHEUTZOW (eds)
354 Groups and analysis, K. TENT (ed)
355 Non-equilibrium statistical mechanics and turbulence, J. CARDY, G. FALKOVICH & K. GAWEDZKI
356 Elliptic curves and big Galois representations, D. DELBOURGO
357 Algebraic theory of differential equations, M.A.H. MACCALLUM & A.V. MIKHAILOV (eds)
358 Geometric and cohomological methods in group theory, M.R. BRIDSON, P.H. KROPHOLLER &
I.J. LEARY (eds)
359 Moduli spaces and vector bundles, L. BRAMBILA-PAZ, S.B. BRADLOW, O. GARCÍA-PRADA &
S. RAMANAN (eds)
360 Zariski geometries, B. ZILBER
361 Words: Notes on verbal width in groups, D. SEGAL
362 Differential tensor algebras and their module categories, R. BAUTISTA, L. SALMERÓN & R. ZUAZUA
363 Foundations of computational mathematics, Hong Kong 2008, F. CUCKER, A. PINKUS & M.J. TODD (eds)
364 Partial differential equations and fluid mechanics, J.C. ROBINSON & J.L. RODRIGO (eds)
365 Surveys in combinatorics 2009, S. HUCZYNSKA, J.D. MITCHELL & C.M. RONEY-DOUGAL (eds)
366 Highly oscillatory problems, B. ENGQUIST, A. FOKAS, E. HAIRER & A. ISERLES (eds)
367 Random matrices: High dimensional phenomena, G. BLOWER
368 Geometry of Riemann surfaces, F.P. GARDINER, G. GONZÁLEZ-DIEZ & C. KOUROUNIOTIS (eds)
369 Epidemics and rumours in complex networks, M. DRAIEF & L. MASSOULIÉ
370 Theory of p-adic distributions, S. ALBEVERIO, A.YU. KHRENNIKOV & V.M. SHELKOVICH
371 Conformal fractals, F. PRZYTYCKI & M. URBAŃSKI
372 Moonshine: The first quarter century and beyond, J. LEPOWSKY, J. MCKAY & M.P. TUITE (eds)
373 Smoothness, regularity, and complete intersection, J. MAJADAS & A. G. RODICIO
374 Geometric analysis of hyperbolic differential equations: An introduction, S. ALINHAC
375 Triangulated categories, T. HOLM, P. JØRGENSEN & R. ROUQUIER (eds)
376 Permutation patterns, S. LINTON, N. RUŠKUC & V. VATTER (eds)
377 An introduction to Galois cohomology and its applications, G. BERHUY
378 Probability and mathematical genetics, N. H. BINGHAM & C. M. GOLDIE (eds)
379 Finite and algorithmic model theory, J. ESPARZA, C. MICHAUX & C. STEINHORN (eds)
380 Real and complex singularities, M. MANOEL, M.C. ROMERO FUSTER & C.T.C WALL (eds)
381 Symmetries and integrability of difference equations, D. LEVI, P. OLVER, Z. THOMOVA &
P. WINTERNITZ (eds)
382 Forcing with random variables and proof complexity, J. KRAJÍČEK
383 Motivic integration and its interactions with model theory and non-Archimedean geometry I, R. CLUCKERS,
J. NICAISE & J. SEBAG (eds)
384 Motivic integration and its interactions with model theory and non-Archimedean geometry II, R. CLUCKERS,
J. NICAISE & J. SEBAG (eds)
385 Entropy of hidden Markov processes and connections to dynamical systems, B. MARCUS, K. PETERSEN &
T. WEISSMAN (eds)
386 Independence-friendly logic, A.L. MANN, G. SANDU & M. SEVENSTER
387 Groups St Andrews 2009 in Bath I, C.M. CAMPBELL et al (eds)
388 Groups St Andrews 2009 in Bath II, C.M. CAMPBELL et al (eds)
389 Random fields on the sphere, D. MARINUCCI & G. PECCATI
390 Localization in periodic potentials, D.E. PELINOVSKY
391 Fusion systems in algebra and topology M. ASCHBACHER, R. KESSAR & B. OLIVER
392 Surveys in combinatorics 2011, R. CHAPMAN (ed)
393 Non-abelian fundamental groups and Iwasawa theory, J. COATES et al (eds)
394 Variational problems in differential geometry, R. BIELAWSKI, K. HOUSTON & M. SPEIGHT (eds)
Conference photograph
London Mathematical Society Lecture Note Series: 394

Variational Problems in
Differential Geometry
University of Leeds 2009

Edited by
R. BIELAWSKI
K. HOUSTON
J.M. SPEIGHT
University of Leeds
 cambridge university press
Cambridge, New York, Melbourne, Madrid, Cape Town,
Singapore, São Paulo, Delhi, Tokyo, Mexico City
Cambridge University Press
The Edinburgh Building, Cambridge CB2 8RU, UK
Published in the United States of America by Cambridge University Press, New York

www.cambridge.org
Information on this title: www.cambridge.org/9780521282741


C Cambridge University Press 2012

This publication is in copyright. Subject to statutory exception

and to the provisions of relevant collective licensing agreements,
no reproduction of any part may take place without the written
permission of Cambridge University Press.

First published 2012

Printed in the United Kingdom at the University Press, Cambridge

A catalogue record for this publication is available from the British Library

Library of Congress Cataloguing in Publication data

Variational problems in differential geometry : University of Leeds, 2009 /
edited by R. Bielawski, K. Houston, J.M. Speight.
p. cm. – (London Mathematical Society lecture note series ; 394)
Includes bibliographical references.
ISBN 978-0-521-28274-1 (pbk.)
1. Geometry, Differential – Congresses. I. Bielawski, R. II. Houston, Kevin, 1968–
III. Speight, J. M. (J. Martin) IV. Title. V. Series.
QA641.V37 2012
516.3 6 – dc23 2011027490

ISBN 978-0-521-28274-1 Paperback

Cambridge University Press has no responsibility for the persistence or

accuracy of URLs for external or third-party internet websites referred to
in this publication, and does not guarantee that any content on such
websites is, or will remain, accurate or appropriate.
 Contents

List of contributors page viii

Preface xi

1 The supremum of first eigenvalues of conformally

covariant operators in a conformal class 1
Bernd Ammann and Pierre Jammes
1.1 Introduction 1
1.2 Preliminaries 4
1.3 Asymptotically cylindrical blowups 11
1.4 Proof of the main theorem 14
Appendix A Analysis on (M∞ , g∞ ) 19
References 22
2 K-Destabilizing test configurations with smooth
central fiber 24
Claudio Arezzo, Alberto Della Vedova, and Gabriele La Nave
2.1 Introduction 24
2.2 The case of normal singularities 29
2.3 Proof of Theorem 2.1.8 and examples 32
References 34
3 Explicit constructions of Ricci solitons 37
Paul Baird
3.1 Introduction 37
3.2 Solitons from a dynamical system 40
3.3 Reduction of the equations to a 2-dimensional system 44
3.4 Higher dimensional Ricci solitons via projection 48
3.5 The 4-dimensional geometry Nil4 50
References 55

v
vi Contents

4 Open Iwasawa cells and applications to surface theory 56

Josef F. Dorfmeister
4.1 Introduction 56
4.2 Basic notation and the Birkhoff decomposition 58
4.3 Iwasawa decomposition 59
4.4 Iwasawa decomposition via Birkhoff decomposition 60
4.5 A function defining the open Iwasawa cells 62
4.6 Applications to surface theory 64
References 66
5 Multiplier ideal sheaves and geometric problems 68
Akito Futaki and Yuji Sano
5.1 Introduction 68
5.2 An overview of multiplier ideal sheaves 72
5.3 Direct relationships between multiplier ideal sheaves and
the obstruction F 83
References 90
6 Multisymplectic formalism and the covariant phase space 94
Frédéric Hélein
6.1 The multisymplectic formalism 95
6.2 The covariant phase space 110
6.3 Geometric quantization 117
References 123
7 Nonnegative curvature on disk bundles 127
Lorenz J. Schwachhöfer
7.1 Introduction 127
7.2 Normal homogeneous metrics and Cheeger deformations 128
7.3 Homogeneous metrics of nonnegative curvature 130
7.4 Collar metrics of nonnegative curvature 131
7.5 Bundles with normal homogeneous collar 132
7.6 Cohomogeneity one manifolds 139
References 140
8 Morse theory and stable pairs 142
Richard A. Wentworth and Graeme Wilkin
8.1 Introduction 142
8.2 Stable pairs 146
8.3 Morse theory 154
8.4 Cohomology of moduli spaces 174
References 180
 Contents vii

9 Manifolds with k-positive Ricci curvature 182

Jon Wolfson
9.1 Introduction 182
9.2 Manifolds with k-positive Ricci curvature 183
9.3 Fill radius and an approach to Conjecture 1 192
9.4 The fundamental group and fill radius bounds 198
References 200
 Contributors

Bernd Ammann
Facultät für Mathematik, Universität Regensburg, 93040 Regensburg,
Germany

Pierre Jammes
Laboratoire J.-A. Dieudonné, Université Nice – Sophia Antipolis, Parc
Valrose, F-06108 NICE Cedex 02, France

Claudio Arezzo
Abdus Salam International Center for Theoretical Physics, Strada Costiera
11, Trieste (Italy) and Dipartimento di Matematica, Università di Parma,
Parco Area delle Scienze 53/A, Parma, Italy

Alberto Della Vedova

Fine Hall, Princeton University, Princeton, NJ 08544 and Dipartimento di
Matematica, Università di Parma, Parco Area delle Scienze 53/A, Parma, Italy

Gabriele La Nave
Department of Mathematics, Yeshiva University, 500 West 185 Street,
New York, NY, USA

Paul Baird
Département de Mathématiques, Université de Bretagne Occidentale,
6 Avenue Le Gorgeu – CS 93837, 29238 Brest, France

Josef F. Dorfmeister
Fakultät für Mathematik, Technische Universität München, Boltzmannstr. 3,
D-85747 Garching, Germany

viii
 List of contributors ix

Akito Futaki
Department of Mathematics, Tokyo Institute of Technology, 2-12-1,
O-okayama, Meguro, Tokyo 152-8551, Japan

Yuji Sano
Department of Mathematics, Kyushu University, 6-10-1, Hakozaki,
Higashiku, Fukuoka-city, Fukuoka 812-8581 Japan

Frédéric Hélein
Institut de Mathématiques de Jussieu, UMR CNRS 7586, Université Denis
Diderot Paris 7, 175 rue du Chevaleret, 75013 Paris, France

Lorenz J. Schwachhöfer
Fakultät für Mathematik, Technische Universität Dortmund, Vogelpothsweg
87, 44221 Dortmund, Germany

Richard A. Wentworth
Department of Mathematics, University of Maryland, College Park, MD
20742, USA

Graeme Wilkin
Department of Mathematics, University of Colorado, Boulder, CO 80309,
USA

Jon Wolfson
Department of Mathematics, Michigan State University, East Lansing,
MI 48824, USA
 Preface

The workshop Variational Problems in Differential Geometry was held at the

University of Leeds from March 30 to April 2nd, 2009.
The aim of the meeting was to bring together researchers working on
disparate geometric problems, all of which admit a variational formulation.
Among the topics discussed were recent developments in harmonic maps and
morphisms, minimal and CMC surfaces, extremal Kähler metrics, the Yam-
abe functional, Hamiltonian variational problems, and topics related to gauge
theory and to the Ricci flow.
The meeting incorporated a special session in honour of John C. Wood, on
the occasion of his 60th birthday, to celebrate his seminal contributions to the
theory of harmonic maps and morphisms.
The following mathematicians gave one-hour talks: Bernd Ammann, Clau-
dio Arezzo, Paul Baird, Olivier Biquard, Christoph Boehm, Francis Burstall,
Josef Dorfmeister, Akito Futaki, Mark Haskins, Frederic Helein, Nicolaos
Kapouleas, Mario Micallef, Frank Pacard, Simon Salamon, Lorenz Schwach-
hoefer, Peter Topping, Richard Wentworth, and Jon Wolfson.
There were about 50 participants from the UK, US, Japan and several Euro-
pean countries. The schedule allowed plenty of opportunities for discussion
and interaction between official talks and made for a successful and stimulat-
ing meeting.
The workshop was financially supported by the London Mathematical Soci-
ety, the Engineering and Physical Sciences Research Council of Great Britain
and the School of Mathematics, University of Leeds.
The articles presented in this volume represent the whole spectrum of the
subject.
The supremum of first eigenvalues of conformally covariant operators in a
conformal class by Ammann and Jammes is concerned with the first eigenvalues
of the Yamabe operator, the Dirac operator, and more general conformally

xi
xii Preface

covariant elliptic operators on compact Riemannian manifolds. It is well known

that the infimum of the first eigenvalue in a given conformal class reflects a rich
geometric structure. In this article, the authors study the supremum of the first
eigenvalue and show that, for a very general class of operators, this supremum
is infinite.
The article, K-Destabilizing test configurations with smooth central fiber
by Arezzo, Della Vedova, and La Nave is concerned with the famous Tian-
Yau-Donaldson conjecture about existence of constant scalar curvature Kähler
metrics. They construct many new families of K-unstable manifolds, and,
consequently, many new examples of manifolds which do not admit Kähler
constant scalar curvature metrics in some cohomology classes.
As has been now understood, a very natural extension of Einstein metrics
are the Ricci solitons. These are the subject of Paul Baird’s article Explicit
constructions of Ricci solitons, in which he does precisely that: he constructs
many explicit examples, including some in the more exotic geometries Sol3 ,
Nil3 , and Nil4 .
Josef Dorfmeister is concerned with a more classical topic: that of constant
mean curvature and Willmore surfaces. In recent years, many new examples of
such surfaces were constructed using loop groups. The method relies on finding
“Iwasawa-like” decompositions of loop groups and the article Open Iwasawa
cells in twisted loop groups and some applications to harmonic maps discusses
such decompositions and their singularities.
The currently extremely important notions of K-stability and K-
polystability are the topic of the paper by Futaki and Sano Multiplier ideal
sheaves and geometric problems. This is an expository article giving state-of-
the-art presentation of the powerful method of multiplier ideal sheaves and
their applications to Kähler-Einstein and Sasaki-Einstein geometries.
Multisymplectic formalism and the covariant phase space by Frédéric Hélein
takes us outside Riemannian geometry. The author presents an alternative (in
fact, two of them) to the Feynman integral as a foundation of quantum field
theory.
Lorenz Schwachhöfer’s Nonnegative curvature on disk bundles is a survey of
the glueing method used to construct Riemannian manifolds with nonnegative
sectional curvature - one of the classical problems in geometry.
Morse theory and stable pairs by Wentworth and Wilkin introduces new
techniques to compute equivariant cohomology of certain natural moduli
spaces. The main ingredient is a version of Morse-Atiyah-Bott theory adapted
to singular infinite dimensional spaces.
The final article, Manifolds with k-positive Ricci curvature, by Jon Wolf-
son, is a survey of results and conjectures about Riemannian n-manifolds with
 Preface xiii

k-positive Ricci curvature. These interpolate between positive scalar curva-

ture (n-positive Ricci curvature) and positive Ricci curvature (1-positive Ricci
curvature), and the author shows how the results about k-positive Ricci curva-
ture, 1 < k < n, also interpolate, or should do, between what is known about
manifolds satisfying those two classical notions of positivity.
We would like to extend our thanks to our colleague John Wood for his help
and assistance in preparing these proceedings.

R. Bielawski
K. Houston
J.M. Speight
Leeds, UK
 1
The supremum of first eigenvalues of
conformally covariant operators
in a conformal class
bernd ammann and pierre jammes

Abstract
Let (M, g) be a compact Riemannian manifold of dimension ≥ 3. We show that
there is a metric g̃ conformal to g and of volume 1 such that the first positive
eigenvalue of the conformal Laplacian with respect to g̃ is arbitrarily large.
A similar statement is proven for the first positive eigenvalue of the Dirac
operator on a spin manifold of dimension ≥ 2.

1.1 Introduction
The goal of this article is to prove the following theorems.

Theorem 1.1.1 Let (M, g0 , χ ) be compact Riemannian spin manifold of

dimension n ≥ 2. For any metric g in the conformal class [g0 ], we denote
the first positive eigenvalue of the Dirac operator on (M, g, χ ) by λ+
1 (Dg ).
Then

sup λ+
1 (Dg )Vol(M, g)
1/n
= ∞.
g∈[g0 ]

Theorem 1.1.2 Let (M, g0 , χ ) be compact Riemannian manifold of dimension

n ≥ 3. For any metric g in the conformal class [g0 ], we denote the first positive
eigenvalue of the conformal Laplacian Lg := g + 4(n−1) n−2
Scalg (also called
+
Yamabe operator) on (M, g, χ ) by λ1 (Lg ). Then

sup λ+
1 (Lg )Vol(M, g)
2/n
= ∞.
g∈[g0 ]

The Dirac operator and the conformal Laplacian belong to a large fam-
ily of operators, defined in details in subsection 1.2.3. These operators are

1
2 B. Ammann and P. Jammes

called conformally covariant elliptic operators of order k and of bidegree

((n − k)/2, (n + k)/2), acting on manifolds (M, g) of dimension n > k. In
our definition we also claim formal self-adjointness.
All such conformally covariant elliptic operators of order k and of bidegree
((n − k)/2, (n + k)/2) share several analytical properties, in particular they are
associated to the non-compact embedding H k/2 → L2n/(n−k) . Often they have
interpretations in conformal geometry. To give an example, we define for a
compact Riemannian manifold (M, g0 )

Y (M, [g0 ]) := inf λ1 (Lg )Vol(M, g)2/n ,

g∈[g0 ]

where λ1 (Lg ) is the lowest eigenvalue of Lg . If Y (M, [g0 ]) > 0, then the
solution of the Yamabe problem [29] tells us that the infimum is attained and
the minimizer is a metric of constant scalar curvature. This famous problem
was finally solved by Schoen and Yau using the positive mass theorem.
In a similar way, for n = 2 the Dirac operator is associated to constant-mean-
curvature conformal immersions of the universal covering into R3 . If a Dirac-
operator-analogue of the positive mass theorem holds for a given manifold
(M, g0 ), then the infimum

inf λ+
1 (Dg )Vol(M, g)
1/n
g∈[g0 ]

is attained [3]. However, it is still unclear whether such a Dirac-operator-

analogue of the positive mass theorem holds in general.
The Yamabe problem and its Dirac operator analogue, as well as the
analogues for other conformally covariant operators are typically solved by
minimizing an associated variational problem. As the Sobolev embedding
H k/2 → L2n/(n−k) is non-compact, the direct method of the calculus of variation
fails, but perturbation techniques and conformal blow-up techniques typically
work. Hence all these operators share many properties.
However, only few statements can be proven simultaneously for all confor-
mally covariant elliptic operators of order k and of bidegree ((n − k)/2, (n +
k)/2). Some of the operators are bounded from below (e.g. the Yamabe and
the Paneitz operator), whereas others are not (e.g. the Dirac operator). Some
of them admit a maximum principle, others do not. Some of them act on func-
tions, others on sections of vector bundles. The associated Sobolev space H k/2
has non-integer order if k is odd, hence it is not the natural domain of a dif-
ferential operator. For Dirac operators, the spin structure has to be considered
in order to derive a statement as Theorem 1.1.1 for n = 2. Because of these
differences, most analytical properties have to be proven for each operator
separately.
 The supremum of first eigenvalues 3

We consider it hence as remarkable that the proof of our Theorems 1.1.1

and 1.1.2 can be extended to all such operators. Our proof only uses some few
properties of the operators, defined axiomatically in 1.2.3. More exactly we
prove the following.

Theorem 1.1.3 Let Pg be a conformally covariant elliptic operator of order

k, of bidegree ((n − k)/2, (n + k)/2) acting on manifolds of dimension n > k.
We also assume that Pg is invertible on Sn−1 × R (see Definition 1.2.4). Let
(M, g0 ) be compact Riemannian manifold. In the case that Pg depends on
the spin structure, we assume that M is oriented and is equipped with a spin
structure. For any metric g in the conformal class [g0 ], we denote the first
positive eigenvalue of Pg by λ+1 (Pg ). Then

sup λ+
1 (Pg )Vol(M, g)
k/n
= ∞.
g∈[g0 ]

The interest in this result is motivated by three questions. At first, as already

mentioned above the infimum

inf λ+
1 (Dg )Vol(M, g)
1/n
g∈[g0 ]

reflects a rich geometrical structure [3], [4], [5], [7], [8], similarly for the
conformal Laplacian. It seems natural to study the supremum as well.
The second motivation comes from comparing Theorem 1.1.3 to results
g
about some other differential operators. For the Hodge Laplacian p acting
g
on p-forms, we have supg∈[g0 ] λ1 (p )Vol(M, g) = +∞ for n ≥ 4 and 2 ≤
2/n

p ≤ n − 2 ([19]). On the other hand, for the Laplacian g acting on functions,

we have

sup λk (g )Vol(M, g)2/n < +∞

g∈[g0 ]

(the case k = 1 is proven in [20] and the general case in [27]). See [25] for a
synthetic presentation of this subject.
The essential idea in our proof is to construct metrics with longer and longer
cylindrical parts. We will call this an asymptotically cylindrical blowup. Such
metrics are also called Pinocchio metrics in [2, 6]. In [2, 6] the behavior of Dirac
eigenvalues on such metrics has already been studied partially, but the present
article has much stronger results. To extend these existing results provides the
third motivation.
Acknowledgments We thank B. Colbois, M. Dahl, and E. Humbert for
many related discussions. We thank R. Gover for some helpful comments on
conformally covariant operators, and for several references. The first author
4 B. Ammann and P. Jammes

wants to thank the Albert Einstein institute at Potsdam-Golm for its very kind
hospitality which enabled to write the article.

1.2 Preliminaries
1.2.1 Notations
In this article By (r) denotes the ball of radius r around y, Sy (r) = ∂By (r)
its boundary. The standard sphere S0 (1) ⊂ Rn in Rn is denoted by Sn−1 , its
volume is ωn−1 . For the volume element of (M, g) we use the notation dv g . In
our article, (V ) (resp. c (V )) always denotes the set of all smooth sections
(resp. all compactly supported smooth sections) of the vector bundle V → M.
For sections u of V → M over a Riemannian manifold (M, g) the Sobolev
norms L2 and H s , s ∈ N, are defined as

u 2L2 (M,g) := |u|2 dv g
M
u 2
H s (M,g) := u 2
L2 (M,g) + ∇u 2
L2 (M,g) + · · · + ∇s u 2
L2 (M,g) .

The vector bundle V will be suppressed in the notation. If M and g

are clear from the context, we write just L2 and H s . The completion of
{u ∈ (V ) | u H s (M,g) < ∞} with respect to the H s (M, g)-norm is denoted
by H s (M,g) (V ), or if (M, g) or V is clear from the context, we alternatively
write H s (V ) or H s (M, g) for H s (M,g) (V ). The same definitions are used for
L2 instead of H s . And similarly C k (M,g) (V ) = C k (V ) = C k (M, g) is the set
of all C k -sections, k ∈ N ∪ {∞}.

1.2.2 Removal of singularities

In the proof we will use the following removal of singularities lemma.

Lemma 1.2.1 (Removal of singularities lemma) Let  be a bounded open

subset of Rn containing 0. Let P be an elliptic differential operator of order k
on , f ∈ C ∞ (), and let u ∈ C ∞ ( \ {0}) be a solution of

Pu = f (1.1)

on  \ {0} with
 
−k
lim |u|r = 0 and lim |u| = 0 (1.2)
ε→0 B (2ε)−B (ε) ε→0 B (ε)
0 0 0
 The supremum of first eigenvalues 5

where r is the distance to 0. Then u is a (strong) solution of (1.1) on . The

same result holds for sections of vector bundles over relatively compact open
subset of Riemannian manifolds.

Proof We show that u is a weak solution of (1.1) in the distributional sense, and
then it follows from standard regularity theory, that it is also a strong solution.
This means that we have to show that for any given compactly supported smooth
test function ψ :  → R we have
 
uP ∗ ψ = f ψ.
 

Let η :  → [0, 1] be a test function that is identically 1 on B0 (ε), has

support in B0 (2ε), and with |∇ m η| ≤ Cm /εm . It follows that

sup |P ∗ (ηψ)| ≤ C(P , , ψ)ε−k ,

on B0 (2ε) \ B0 (ε) and sup |P ∗ (ηψ)| ≤ C(P , , ψ) on B0 (ε) and hence

   
 
 uP ∗ (ηψ) ≤ Cε−k |u| + C |u|
 
 B0 (2ε)\B0 (ε) B0 (ε)
  (1.3)
≤C |u|r −k + C |u| → 0.
B0 (2ε)\B0 (ε) B0 (ε)

We conclude
  
uP ∗ ψ = uP ∗ (ηψ) + uP ∗ ((1 − η)ψ)

 
 

= ∗
uP (ηψ) + (P u)(1 − η)ψ (1.4)
 

  
 
→0 →  fψ

for ε → 0. Hence the lemma follows. 


Condition (1.2) is obviously satisfied if  |u|r −k < ∞. It is also satisfied if

|u|2 r −k < ∞ and k ≤ n, (1.5)


as in this case
 2  
−k 2 −k
|u|r ≤ |u| r r −k .
B0 (2ε)\B0 (ε)  B0 (2ε)\B0 (ε)
  
≤C
6 B. Ammann and P. Jammes

1.2.3 Conformally covariant elliptic operators

In this subsection we present a class of certain conformally covariant elliptic
operators. Many important geometric operators are in this class, in particular
the conformal Laplacian, the Paneitz operator, the Dirac operator, see also
[18, 21, 22] for more examples. Readers who are only interested in the Dirac
operator, the Conformal Laplacian or the Paneitz operator, can skip this part
and continue with section 1.3.
Such a conformally covariant operator is not just one single differential oper-
ator, but a procedure how to associate to an n-dimensional Riemannian manifold
(M, g) (potentially with some additional structure) a differential operator Pg
of order k acting on a vector bundle. The important fact is that if g2 = f 2 g1 ,
then one claims

Pg2 = f −
n+k n−k
2 Pg1 f 2 . (1.6)

One also expresses this by saying that P has bidegree ((n − k)/2, (n + k)/2).
The sense of this equation is apparent if Pg is an operator from C ∞ (M)
to C ∞ (M). If Pg acts on a vector bundle or if some additional structure (as
e.g. spin structure) is used for defining it, then a rigorous and careful defini-
tion needs more attention. The language of categories provides a good formal
framework [30]. The concept of conformally covariant elliptic operators is
already used by many authors, but we do not know of a reference where a
formal definition is carried out that fits to our context. (See [26] for a similar
categorial approach that includes some of the operators presented here.) Often
an intuitive definition is used. The intuitive definition is obviously sufficient if
one deals with operators acting on functions, such as the conformal Laplacian
or the Paneitz operator. However to properly state Theorem 1.1.3 we need the
following definition.
Let Riemn (resp. Riemspinn ) be the category n-dimensional Riemannian
manifolds (resp. n-dimensional Riemannian manifolds with orientation and
spin structure). Morphisms from (M1 , g1 ) to (M2 , g2 ) are conformal embed-
dings (M1 , g1 ) → (M2 , g2 ) (resp. conformal embeddings preserving orienta-
tion and spin structure).
Let Laplacenk (resp. Diracnk ) be the category whose objects are

{(M, g), Vg , Pg }

where (M, g) in an object of Riemn (resp. Riemspinn ), where Vg is a vector

bundle with a scalar product on the fibers, where Pg : (Vg ) → (Vg ) is an
elliptic formally self-adjoint differential operator of order k.
 The supremum of first eigenvalues 7

A morphism (ι, κ) from {(M1 , g1 ), Vg1 , Pg1 } to {(M2 , g2 ), Vg2 , Pg2 } consists
of a conformal embedding ι : (M1 , g1 ) → (M2 , g2 ) (preserving orientation
and spin structure in the case of Diracnk ) together with a fiber isomorphism
κ : ι∗ Vg2 → Vg1 preserving fiberwise length, such that Pg1 and Pg2 sat-
isfy the conformal covariance property (1.6). For stating this property pre-
cisely, let f > 0 be defined by ι∗ g2 = f 2 g1 , and let κ∗ : (Vg2 ) → (Vg1 ),
κ∗ (ϕ) = κ ◦ ϕ ◦ ι. Then the conformal covariance property is

κ∗ Pg2 = f −
n+k n−k
2 Pg1 f 2 κ∗ . (1.7)

In the following the maps κ and ι will often be evident from the context
and then will be omitted. The transformation formula (1.7) then simplifies
to (1.6).

Definition 1.2.2 A conformally covariant elliptic operator of order k and of

bidegree ((n − k)/2, (n + k)/2) is a contravariant functor from Riemn (resp.
Riemspinn ) to Laplacenk (resp. Diracnk ), mapping (M, g) to (M, g, Vg , Pg ) in
such a way that the coefficients are continuous in the C k -topology of metrics
(see below). To shorten notation, we just write Pg or P for this functor.

It remains to explain the C k -continuity of the coefficients.

For Riemannian metrics g, g1 , g2 defined on a compact set K ⊂ M we set
g
dC k (K) (g1 , g2 ) := max (∇g )t (g1 − g2 ) C 0 (K) .
t=0,...,k

g
For a fixed background metric g, the relation dC k (K) ( · , · ) defines a distance
function on the space of metrics on K. The topology induced by d g is inde-
pendent of this background metric and it is called the C k -topology of metrics
on K.

Definition 1.2.3 We say that the coefficients of P are continuous in the C k -

topology of metrics if for any metric g on a manifold M, and for any compact
subset K ⊂ M there is a neighborhood U of g|K in the C k -topology of met-
rics on K, such that for all metrics g̃, g̃|K ∈ U, there is an isomorphism of
vector bundles κ̂ : Vg |K → Vg̃ |K over the identity of K with induced map
κ̂∗ : (Vg |K ) → (Vg̃ |K ) with the property that the coefficients of the differ-
ential operator

Pg − (κ̂∗ )−1 Pg̃ κ̂∗

depend continuously on g̃ (with respect to the C k -topology of metrics).

8 B. Ammann and P. Jammes

1.2.4 Invertibility on Sn−1 × R

Let P be a conformally covariant elliptic operator of order k and of bide-
gree ((n − k)/2, (n + k)/2). For (M, g) = Sn−1 × R, the operator Pg is a
self-adjoint operator H k ⊂ L2 → L2 (see Lemma 1.3.1 and the comments
thereafter).
Definition 1.2.4 We say that P is invertible on Sn−1 × R if Pg is an invertible
operator H k → L2 where g is the standard product metric on Sn−1 × R. In
order words there is a constant σ > 0 such that the spectrum of Pg : H k (Vg ) →
L2 (Vg ) is contained in (−∞, −σ ] ∪ [σ, ∞) for any g ∈ U . In the following,
the largest such σ will be called σP .
We conjecture that any conformally covariant elliptic operator of order k
and of bidegree ((n − k)/2, (n + k)/2) with k < n is invertible on Sn−1 × R.

1.2.5 Examples
Example 1: The Conformal Laplacian
Let
n−2
Lg := g + Scalg ,
4(n − 1)
be the conformal Laplacian. It acts on functions on a Riemannian manifold
(M, g), i.e. Vg is the trivial real line bundle R. Let ι : (M1 , g1 ) → (M2 , g2 )
be a conformal embedding. Then we can choose κ := Id : ι∗ Vg2 → Vg1 and
formula (1.7) holds for k = 2 (see e.g. [15, Section 1.J]). All coefficients of
Lg depend continuously on g in the C 2 -topology. Hence L is a conformally
covariant elliptic operator of order 2 and of bidegree ((n − 2)/2, (n + 2)/2).
The scalar curvature of Sn−1 × R is (n − 1)(n − 2). The spectrum of Lg on
Sn−1 × R of Lg coincides with the essential spectrum of Lg and is [σL , ∞) with
σL := (n − 2)2 /4. Hence L is invertible on S n−1 × R if (and only if) n > 2.
Example 2: The Paneitz operator
Let (M, g) be a smooth, compact Riemannian manifold of dimension n ≥ 5.
The Paneitz operator Pg is given by
n−4
Pg u = (g )2 u − divg (Ag du) + Qg u
2
where
(n − 2)2 + 4 4
Ag := Scalg g − Ricg ,
2(n − 1)(n − 2) n−2

1 n3 − 4n2 + 16n − 16 2
Qg = g Scalg + Scal2g − |Ricg |2 .
2(n − 1) 8(n − 1) (n − 2)
2 2 (n − 2)2
 The supremum of first eigenvalues 9

This operator was defined by Paneitz [32] in the case n = 4, and it was general-
ized by Branson in [17] to arbitrary dimensions ≥ 4. We also refer to Theorem
1.21 of the overview article [16]. The explicit formula presented above can
be found e.g. in [23]. The coefficients of Pg depend continuously on g in the
C 4 -topology
As in the previous example we can choose for κ the identity, and then the
Paneitz operator Pg is a conformally covariant elliptic operator of order 4 and
of bidegree ((n − 4)/2, (n + 4)/2).
On Sn−1 × R one calculates
(n − 4)n
Ag = Id + 4πR > 0
2
where πR is the projection to vectors parallel to R.

(n − 4)n2
Qg = .
8
We conclude
(n − 4)n2
σP = Q =
8
and P is invertible on Sn−1 × R if (and only if) n > 4.
Examples 3: The Dirac operator.
Let g̃ = f 2 g. Let g M resp. g̃ M be the spinor bundle of (M, g) resp.
g
(M, g̃). Then there is a fiberwise isomorphism βg̃ : g M → g̃ M, preserving
the norm such that

Dg̃ ◦ βg̃ (ϕ) = f −

g n+1 g n−1
2 βg̃ ◦ Dg f 2 ϕ ,

see [24, 14] for details. Furthermore, the cocycle conditions

g g̃ g
βg̃ ◦ βgg̃ = Id and βgĝ ◦ βĝ ◦ βg̃ = Id
g
hold for conformal metrics g, g̃ and ĝ. We will hence use the map βg̃ to identify
g M with g̃ M. Hence we simply get

Dg̃ ϕ = f −
n+1 n−1
2 ◦ Dg f 2 ϕ . (1.8)

All coefficients of Dg depend continuously on g in the C 1 -topology. Hence

D is a conformally covariant elliptic operator of order 1 and of bidegree
((n − 1)/2, (n + 1)/2).
The Dirac operator on Sn−1 × R can be decomposed in a part Dvert deriving
along Sn−1 and a part Dhor deriving along R, Dg = Dvert + Dhor , see [1] or [2].
10 B. Ammann and P. Jammes

Locally
n−1
Dvert = ei · ∇ei
i=1

for a local frame (e1 , . . . , en−1 ) of Sn−1 . Here · denotes the Clifford multi-
plication T M ⊗ g M → g M. Furthermore Dhor = ∂t · ∇∂t , where t ∈ R is
the standard coordinate of R. The operators Dvert and Dhor anticommute. For
n ≥ 3, the spectrum of Dvert coincides with the spectrum of the Dirac operator
on Sn−1 , we cite [12] and obtain
 
n−1
specDvert = ± + k | k ∈ N0 .
2

The operator (Dhor )2 is the ordinary Laplacian on R and hence has spectrum
[0, ∞). Together this implies that the spectrum of the Dirac operator on Sn−1 ×
R is the set (−∞, −σD ] ∪ [σD , ∞) with σD = n−1 2
.
In the case n = 2 these statements are only correct if the circle Sn−1 = S1
carries the spin structure induced from the ball. Only this spin structure extends
to the conformal compactification that is given by adding one point at infinity
for each end. For this reason, we will understand in the whole article that all
circles S1 should be equipped with this bounding spin structure. The exten-
sion of the spin structure is essential in order to have a spinor bundle on the
compactification. The methods used in our proof use this extension implicitly.
Hence D is invertible on S n−1 × R if (and only if) n > 1.
Most techniques used in the literature on estimating eigenvalues of the
Dirac operators do not use the spin structure and hence these techniques cannot
provide a proof in the case n = 2.
Example 4: The Rarita-Schwinger operator and many other Fegan type
operators are conformally covariant elliptic operators of order 1 and of bide-
gree ((n − 1)/2, (n + 1)/2). See [21] and in the work of T. Branson for more
information.
Example 5: Assume that (M, g) is a Riemannian spin manifold that carries
a vector bundle W → M with metric and metric connection. Then there is a
natural first order operator (g M ⊗ W ) → (g M ⊗ W ), the Dirac opera-
tor twisted by W . This operator has similar properties as conformally covariant
elliptic operators of order 1 and of bidegree ((n − 1)/2, (n + 1)/2). The meth-
ods of our article can be easily adapted in order to show that Theorem 1.1.3
is also true for this twisted Dirac operator. However, twisted Dirac operators
are not “conformally covariant elliptic operators” in the above sense. They
could have been included in this class by replacing the category Riemspinn by
 The supremum of first eigenvalues 11

Figure 1.1 Asymptotically cylindrical metrics gL (alias Pinocchio metrics) with

growing nose length L.

a category of Riemannian spin manifolds with twisting bundles. In order not to

overload the formalism we chose not to present these larger categories.
The same discussion applies to the spinc -Dirac operator of a spinc -manifold.

1.3 Asymptotically cylindrical blowups

1.3.1 Convention
From now on we suppose that Pg is a conformally covariant elliptic operator of
order k, of bidegree ((n − k)/2, (n + k)/2), acting on manifolds of dimension
n and invertible on Sn−1 × R.

1.3.2 Definition of the metrics

Let g0 be a Riemannian metric on a compact manifold M. We can suppose
that the injectivity radius in a fixed point y ∈ M is larger than 1. The geodesic
distance from y to x is denoted by d(x, y).
We choose a smooth function F∞ : M \ {y} → [1, ∞) such such that
F∞ (x) = 1 if d(x, y) ≥ 1, F∞ (x) ≤ 2 if d(x, y) ≥ 1/2 and such that F∞ (x) =
d(x, y)−1 if d(x, y) ∈ (0, 1/2]. Then for L ≥ 1 we define FL to be a smooth
positive function on M, depending only on d(x, y), such that FL (x) = F∞ (x)
if d(x, y) ≥ e−L and FL (x) ≤ d(x, y)−1 = F∞ (x) if d(x, y) ≤ e−L .
For any L ≥ 1 or L = ∞ set gL := FL2 g0 . The metric g∞ is a complete
metric on M∞ .
The family of metrics (gL ) is called an asymptotically cylindrical blowup,
in the literature it is denoted as a family of Pinocchio metrics [6], see also
Figure 1.1.

1.3.3 Eigenvalues and basic properties on (M, gL )

For the P -operator associated to (M, gL ), L ∈ {0} ∪ [1, ∞) (or more exactly
its self-adjoint extension) we simply write PL instead of PgL . As M is compact
the spectrum of PL is discrete.
12 B. Ammann and P. Jammes

We will denote the spectrum of PL in the following way

. . . ≤ λ− + +
1 (PL ) < 0 = 0 . . . = 0 < λ1 (PL ) ≤ λ2 (PL ) ≤ . . . ,

where each eigenvalue appears with the multiplicity corresponding to the

dimension of the eigenspace. The zeros might appear on this list or not, depend-
ing on whether PL is invertible or not. The spectrum might be entirely positive
(for example the conformal Laplacian Yg on the sphere) in which case λ− 1 (PL )
is not defined. Similarly, λ+
1 (P L ) is not defined if the spectrum of (PL ) is
negative.

1.3.4 Analytical facts about (M∞ , g∞ )

The analysis of non-compact manifolds as (M∞ , g∞ ) is more complicated than
in the compact case. Nevertheless (M∞ , g∞ ) is an asymptotically cylindrical
manifold, and for such manifolds an extensive literature is available. One pos-
sible approach would be Melrose’s b-calculus [31]: our cylindrical manifold is
such a b-manifold, but for simplicity and self-containedness we avoid this the-
ory. We will need some few properties that we will summarize in the following
proposition.
We assume in the whole section that P is a conformally covariant elliptic
operator that is invertible on Sn−1 × R, and we write P∞ := Pg∞ for the operator
acting on sections of the bundle V over (M∞ , g∞ ).

Proposition 1.3.1 P∞ extends to a bounded operator from

H k (M∞ ,g∞ ) (V ) → L2 (M∞ ,g∞ ) (V )

and it satisfies the following regularity estimate

(∇ ∞ )s u L2 (M∞ ,g∞ ) ≤ C( u L2 (M∞ ,g∞ ) + P∞ u L2 (M∞ ,g∞ ) ) (1.9)

for all u ∈ H k (M∞ ,g∞ ) (V ) and all s ∈ {0, 1, . . . , k}. The operator

P∞ : H k (M∞ ,g∞ ) (V ) → L2 (M∞ ,g∞ ) (V )

is self-adjoint in the sense of an operator in L2 (M∞ ,g∞ ) (V ).

The proof of the proposition will be sketched in the appendix.

Proposition 1.3.2 The essential spectrum of P∞ coincides with the essen-

tial spectrum of the P -operator on the standard cylinder Sn−1 × R. Thus the
essential spectrum of P∞ is contained in (−∞, −σP ] ∪ [σP , ∞).
 The supremum of first eigenvalues 13

This proposition follows from the characterization of the essential spectrum

in terms of Weyl sequences, a well-known technique which is for example
carried out and well explained in [13].
The second proposition states that the spectrum of P∞ in the interval
(−σP , σP ) is discrete as well. Eigenvalues of P∞ in this interval will be called
small eigenvalues of P∞ . Similarly to above we use the notation λ± j (P∞ ) for
the small eigenvalues of P∞ .

1.3.5 The kernel

Having recalled these well-known facts we will now study the kernel of con-
formally covariant operators.
If g and g̃ = f 2 are conformal metrics on a compact manifold M, then

ϕ → f −
n−k
2 ϕ

obviously defines an isomorphism from ker Pg to ker Pg̃ . It is less obvious that
a similar statement holds if we compare g0 and g∞ defined before:

Proposition 1.3.3 The map

ker P0 → ker P∞
− n−k
ϕ0 → ϕ∞ = F∞ 2
ϕ0

is an isomorphism of vector spaces.

Proof Suppose ϕ0 ∈ ker P0 . Using standard regularity results it is clear that

sup |ϕ0 | < ∞. Then
  
−(n−k)
|ϕ∞ |2 dv g∞ ≤ |ϕ∞ |2 dv g∞ + sup |ϕ0 |2 F∞ dv g∞
M∞ M\By (1/2) By (1/2)
  1/2 n−1
r
≤2 k
|ϕ0 | dv + sup |ϕ0 | ωn−1
2 g0 2
dr < ∞.
M\By (1/2) 0 rk
(1.10)

Here we used that up to lower order terms dv g∞ coincides with the product
measure of the standard measure on the sphere with the measure d(log r) =
1
r
dr. Furthermore, formula (1.6) implies P∞ ϕ∞ = 0. Hence the map is well-
defined. In order to show that it is an isomorphism we show that the obvious
n−k
inverse ϕ∞ → ϕ0 := F∞2 ϕ∞ is well defined. To see this we start with an
L2 -section in the kernel of P∞ .
14 B. Ammann and P. Jammes

We calculate
 
k
F∞ |ϕ0 |2 dv g0
= |ϕ∞ |2 dv g∞ .
M M∞

Using again (1.6) we see that this section satisfies P0 ϕ0 = 0 on M \ {y}. Hence
condition (1.5) is satisfied, and together with the removal of singularity lemma
(Lemma 1.2.1) one obtains that the inverse map is well defined. The proposition
follows. 

1.4 Proof of the main theorem

1.4.1 Stronger version of the main theorem
We will now show the following theorem.

Theorem 1.4.1 Let P be a conformally covariant elliptic operator of order

k, of bidegree ((n − k)/2, (n + k)/2), on manifolds of dimension n > k. We
assume that P is invertible on Sn−1 × R.
If lim inf L→∞ |λ±
j (PL )| < σP , then

λ± ±
j (PL ) → λj (P∞ ) ∈ (−σP , σP ) for L → ∞.

In the case Spec(Pg0 ) ⊂ (0, ∞) the theorem only makes a statement about
λ+j and conversely in the case that Spec(Pg0 ) ⊂ (−∞, 0) it only makes a
,
statement about λ−
j .
Obviously this theorem implies Theorem 1.1.3.

1.4.2 The supremum part of the proof of Theorem 1.4.1

At first we prove that

lim sup(λ+ +
j (PL )) ≤ λj (P∞ ). (1.11)
L→∞

Let ϕ1 , . . . , ϕj be sequence of L2 -orthonormal eigenvectors of P∞ to

eigenvalues λ+ +
1 (P∞ ), . . . , λj (P∞ ) ∈ [−λ̄, λ̄], λ̄ < σP . We choose a cut-off
function χ : M → [0, 1] with χ (x) = 1 for − log(d(x, y)) ≤ T , χ (y) = 0
for − log(d(x, y)) ≥ 2T , and |(∇ ∞ )s χ |g∞ ≤ Cs /T s for all s ∈ {0, . . . , k}.
Let ϕ be a linear combination of the eigenvectors ϕ1 , . . . , ϕj . From Propo-
sition 1.3.1 we see that

(∇ ∞ )s ϕ L2 (M∞ ,g∞ ) ≤C ϕ L2 (M∞ ,g∞ )

 The supremum of first eigenvalues 15

where C only depends on (M∞ , g∞ ). Hence for sufficiently large T

P∞ (χ ϕ) − χ P∞ ϕ L2 (M∞ ,g∞ ) ≤ kC/T ϕ L2 (M∞ ,g∞ ) ≤ 2kC/T χ ϕ L2 (M∞ ,g∞ )

as χ ϕ L2 (M∞ ,g∞ ) → ϕ L2 (M∞ ,g∞ ) for T → ∞. The section χ ϕ can be inter-

preted as a section on (M, gL ) if L > 2T , and on the support of χ ϕ
we have gL = g∞ and P∞ (χ ϕ) = PL (χ ϕ). Hence standard Rayleigh quo-
tient arguments imply that if P∞ has m eigenvalues (counted with mul-
tiplicity) in the interval [a, b] then PL has m eigenvalues in the interval
[a − 2kC/T , b + 2kC/T ]. Taking the limit T → ∞ we obtain (1.11).
By exchanging some obvious signs we obtain similarly

lim sup(−λ− −
j (PL )) ≤ −λj (P∞ ). (1.12)
L→∞

1.4.3 The infimum part of the proof of Theorem 1.4.1

We now prove

lim inf (±λ± ±

j (PL )) ≥ ±λj (P∞ ). (1.13)
L→∞

We assume that we have a sequence Li → ∞, and that for each i we have a

system of orthogonal eigenvectors ϕi,1 , . . . , ϕi,m of PLi , i.e. PLi ϕi, = λi, ϕi,
for  ∈ {1, . . . , m}. Furthermore we suppose that λi, → λ̄ ∈ (−σP , σP ) for
 ∈ {1, . . . , m}.
Then
n−k
F Li 2
ψi, := ϕi,
F∞
satisfies
k
F Li
P∞ ψi, = hi, ψi, with hi, := λi, .
F∞
Furthermore
 −k
FLi
ψi, 2L2 (M∞ ,g∞ ) = |ϕi, |2 dv gLi
M F∞
 −k
FLi
≤ sup |ϕi, |2 dv gLi
M M F∞
Because of
 −k 
FLi
dv gL
≤C r n−1−k dr < ∞
M F∞
16 B. Ammann and P. Jammes

(for n > k) the norm ψi, L2 (M∞ ,g∞ ) is finite as well, and we can renormalize
such that

ψi, L2 (M∞ ,g∞ ) = 1.

Lemma 1.4.2 For any δ > 0 and any  ∈ {0, . . . , m} the sequence
 
ψi, C k+1 (M\By (δ),g∞ ) i

is bounded.

Proof of the lemma. After removing finitely many i, we can assume that λi ≤
2λ̄ and e−Li < δ/2. Hence FL = F∞ and hi = λi on M \ By (δ/2). Because of
 
|(P∞ ) ψi | dv
s 2 g∞
≤ (2λ̄) 2s
|ψi |2 dv g∞ ≤ (2λ̄)2s
M\By (δ/2) M\By (δ/2)

we obtain boundedness of ψi in the Sobolev space H sk (M \ By (3δ/4), g∞ ),

and hence, for sufficiently large s boundedness in C k+1 (M \ By (δ), g∞ ). The
lemma is proved. 
Hence after passing to a subsequence ψi, converges in C k,α (M \ By (δ), g∞ )
to a solution ψ̄ of

P∞ ψ̄ = λ̄ ψ̄ .

k,α
By taking a diagonal sequence, one can obtain convergence in Cloc (M∞ ) of
ψi, to ψ̄ . It remains to prove that ψ̄1 , . . . ,ψ̄m are linearly independent, in
particular that any ψ̄ = 0. For this we use the following lemma.

Lemma 1.4.3 For any ε > 0 there is δ0 and i0 such that

ψi, L2 (By (δ0 ),g∞ ) ≤ ε ψi, L2 (M∞ ,g∞ )

for all i ≥ i0 and all  ∈ {0, . . . , m}. In particular,

ψi, L2 (M\By (δ0 ),g∞ ) ≥ (1 − ε) ψi, L2 (M∞ ,g∞ ) .

Proof of the lemma. Because of Proposition 1.3.1 and

P∞ ψi, L2 (M∞ ,g∞ ) ≤ |λ̄ | ψi, L2 (M∞ ,g∞ ) = |λ̄ |

 The supremum of first eigenvalues 17

we get

(∇ ∞ )s ψi, L2 (M∞ ,g∞ ) ≤C

for all s ∈ {0, . . . , k}. Let χ be a cut-off function as in Subsection 1.4.2 with
T = − log δ. Hence
  C C
P∞ (1 − χ )ψi, − (1 − χ )P∞ (ψi, ) L2 (M∞ ,g∞ ) ≤ = . (1.14)
T − log δ

On the other hand (By (δ) \ {y}, g∞ ) converges for suitable choices of base
points for δ → 0 to Sn−1 × (0, ∞) in the C ∞ -topology of Riemannian man-
ifolds with base points. Hence there is a function τ (δ) converging to 0 such
that
 
P∞ (1 − χ )ψi, L2 (M∞ ,g∞ ) ≥ (σp − τ (δ)) (1 − χ )ψi, L2 (M∞ ,g∞ ) . (1.15)

Using the obvious relation

(1 − χ )P∞ (ψi, ) L2 (M∞ ,g∞ ) ≤ |λi, | (1 − χ )ψi, L2 (M∞ ,g∞ )

we obtain with (1.14) and (1.15)

ψi, L2 (By (δ 2 ),g∞ ) ≤ (1 − χ )ψi, L2 (M∞ ,g∞ )

C
≤ .
| log δ|(σP − τ (δ) − |λi, |)

The right hand side is smaller than ε for i sufficiently large and δ suffi-
ciently small. The main statement of the lemma then follows for δ0 := δ 2 .
The Minkowski inequality yields.

ψi, L2 (M\By (δ 2 ),g∞ ) ≥ 1 − ψi, L2 (By (δ 2 ),g∞ ) ≥ 1 − ε. 

The convergence in C 1 (M \ By (δ0 )) implies strong convergence in L2 (M \
By (δ0 ), g∞ ) of ψi, to ψ̄ . Hence

ψ̄ L2 (M\By (δ0 ),g∞ ) ≥ 1 − ε,

and thus ψ̄ L2 (M∞ ,g∞ ) = 1. The orthogonality of these sections is pro-
vided by the following lemma, and the inequality (1.13) then follows
immediately.

Lemma 1.4.4 The sections ψ̄1 , . . . , ψ̄m are orthogonal.

18 B. Ammann and P. Jammes

Proof of the lemma. The sections ϕi,1 , . . . , ϕi, are orthogonal. For any fixed
δ0 (given by the previous lemma), it follows for sufficiently large i that
   
   
 ψi, , ψi,˜ dv g∞  =  ϕi, , ϕi,˜ dv gLi 
M\By (δ0 ) M\By (δ0 )
 
 
= ϕi, , ϕi,˜ dv gLi 
By (δ0 )
 FLi k  (1.16)
 
= ψi, , ψi,˜ dv g∞ 
By (δ0 ) F∞
  
≤1

≤ ε2
Because of strong L2 convergence on M \ By (δ0 ) this implies
 
 
 ψ̄ , ψ̄˜ dv g∞  ≤ ε2 (1.17)
M\By (δ0 )

for ˜ = , and hence in the limit ε → 0 (and δ0 → 0) we get the orthogonality

of ψ̄1 , . . . , ψ̄m . 
 Appendix A Analysis on (M∞ , g∞ )

The aim of this appendix is to sketch how to prove Proposition 1.3.1. All
properties in this appendix are well-known to experts, but explicit references
are not evident to find. Thus this summary might be helpful to the reader.
The geometry of (M∞ , g∞ ) is asymptotically cylindrical. The metric g∞
is even a b-metric in the sense of Melrose [31], but to keep the presentation
simple, we avoid the b-calculus.
If (r, γ ) ∈ R+ × Sn−1 denote polar normal coordinates with respect to the
metric g0 , and if we set t := − log r, then (t, γ ) defines a diffeomorphism α :
By 0 (1/2) \ {y} → [log 2, ∞) × Sn−1 such that (α −1 )∗ g∞ = dt 2 + ht for a
(M,g )

family of metrics such that (α −1 )∗ g∞ , all of its derivatives, its curvature, and all
derivatives of the curvature tend to the standard metric on the cylinder, and the
speed of the convergence is majorised by a multiple of et . Thus the continuity
of the coefficients property implies, that P∞ extends to a bounded operator
from H k (M∞ ,g∞ ) (V ) → L2 (M∞ ,g∞ ) (V ).
The formal self-adjointness of P∞ implies that
 
ψ, P∞ ϕ = P∞ ψ, ϕ (A.18)
M∞ M∞

holds for ϕ, ψ ∈ c (V ) and as c (V ) is dense in H k , property (A.18) follows

all H k -sections ϕ, ψ.
To show Proposition 1.3.1 it remains to prove the regularity estimate and
then to verify that the adjoint of P∞ : H k (M∞ ,g∞ ) (V ) → L2 (M∞ ,g∞ ) (V ) has
domain H k (M∞ ,g∞ ) (V ).
For proving the regularity estimate we need the following local estimate.

Lemma A.1 Let K be a compact subset of a Riemannian manifold (U, g).

Let P be an elliptic differential operator on U of order k ≥ 1. Then there is a

19
20 B. Ammann and P. Jammes

constant C = C(U, K, P , g) such that

 
u H k (K,g) ≤ C u L2 (U,g) + P u L2 (U,g) . (A.19)
k
Here the H (K, g)-norm is defined via the Levi-Civita connection for g.
This estimate holds uniformly in an ε-neighborhood of P and g in the
following sense. Assume that P̃ is another differential operator, and that the
C 0 -norm of the coeffcients of P̃ − P is at most ε, where ε is small. Also
assume that g̃ is ε-close to g in the C k -topology. Then the estimate (A.19)
holds for P̃ instead of P and for g̃ instead of g and again for a constant
C = C(U, K, P , g, ε).

Proof of the lemma. We cover the compact set K by a finite number of

coordinate neighborhoods U1 , . . . , Um . We choose open sets Vi ⊂ Ui such that
the closure of Vi is compact in Ui and such that K ⊂ V1 ∪ . . . ∪ Vm . One can
choose compact sets Ki ⊂ Vi such that K = K1 ∪ . . . ∪ Km . To prove (A.19)
it is sufficient to prove u H k (Ki ,g) ≤ C( u L2 (Vi ,g) + P u L2 (Vi ,g) ) for any i.
We write this inequality in coordinates. As the closure of Vi is a compactum
in Ui , the transition to coordinates changes the above inequality only by a
constant. The operator P , written in a coordinate chart is again elliptic.
We have thus reduced the prove of (A.19) to the prove of the special case
that U and K are open subsets of flat Rn .
The proof of this special case is explained in detail for example in in [33,
Corollary III 1.5]. The idea is to construct a parametrix for P , i.e. a pseudodif-
ferential operator of order −k such that S1 := QP − Id and S2 := P Q − Id are
infinitely smoothing operators. Thus Q is bounded from L2 (U ) to the Sobolev
space H k (U ), in particular Q(P (u)) H k ≤ C P (u) L2 . Smoothing operators
map the Sobolev space L2 continuously to H k . We obtain
u H k (K) ≤ u H k (U ) ≤ Q(P (u)) H k (U ) + S1 (u) H k (U )
 
≤ C  P (u) L2 (U ) + u L2 (U ) .

See also [28, III §3] for a good presentation on how to construct and work with
such a parametrix.
To see the uniformicity, one verifies that
 
 u H k (K,g̃) 
 − 1 ≤ C g̃ − g C k ≤ Cε
 u k
H (K,g)

and
 
 P̃ (u) 
 L2 (U )
− 1 ≤ Cε u H k (U ) .
 P (u)
L2 (U )

The unformicity statement thus follows. 

 The supremum of first eigenvalues 21

Proof of the regularity estimate in Proposition 1.3.1. We write M∞ as MB ∪

([0, ∞) × S n−1 ), such that the metric g∞ is asymptotic (in the C ∞ -sense) to the
standard cylindrical metric. The metric g∞ restricted to [R − 1, R + 2] × S n−1
then converges in the C k -topology to the cylindrical metric dt 2 + σ n−1 on
[0, 3] × S n−1 for R → ∞. As the coefficients of Pg depend continuously on
the metric, the P -operators on [R − 1, R + 2] × S n−1 is in an ε-neighborhood
of P , for R ≥ R0 = R0 (ε). Applying the preceding lemma for K = [R, R +
1] × S n−1 and U = (R − 1, R + 2) × S n−1 we obtain

∇s u L2 ([R,R+1]×S n−1 ,g∞ ) ≤C u L2 ((R−1,R+2)×S n−1 ,g∞ )

+ P∞ u L2 ((R−1,R+2)×S n−1 ,g∞ ) . (A.20)

Similarly, applying the lemma to K = MB ∪ ([0, R0 ] × S n−1 ) and U =

MB ∪ ([0, R0 + 1) × S n−1 ) gives

∇s u L2 (MB ∪([0,R0 ]×S n−1 ),g∞ ) ≤C u L2 (MB ∪([0,R0 +1)×S n−1 ),g∞ )

+ P∞ u L2 (MB ∪([0,R0 +1)×S n−1 ),g∞ ) . (A.21)

Taking the sum of estimate (A.21), of estimate (A.20) for R = R0 , again

estimate (A.20) but for R = R0 + 1, and so for all R ∈ {R0 + 2, R0 + 3, . . .}
we obtain (1.9), with a larger constant C. 
Now we study the domain D of the adjoint of

P∞ : H k (M∞ ,g∞ ) (V ) → L2 (M∞ ,g∞ ) (V ).

By definition a section ϕ : L2 (M∞ ,g∞ ) (V ) is in D if and only if


H k (M∞ ,g∞ ) (V )  u → P∞ u, ϕ (A.22)
M∞

is bounded as a map from L2 to R. For ϕ ∈ H k (M∞ ,g∞ ) (V ) we know that

P∞ ϕ is L2 and thus property (A.18) directly implies this boundedness. Thus
H k (M∞ ,g∞ ) (V ) ⊂ D.
Conversely assume the boundedness  of (A.22). Then there is a v ∈
L2 (M∞ ,g∞ ) (V ) such that M∞ u, v = M∞ P∞ u, ϕ, or in other words P∞ ϕ =
v holds weakly. Standard regularity theory implies

ϕ ∈ H k (M∞ ,g∞ ) (V ).

We obtain H k (M∞ ,g∞ ) (V ) = D, and thus the self-adjointness of P∞ follows.

Proposition 1.3.1 is thus shown.
22 B. Ammann and P. Jammes

References
[1] B. Ammann, The Dirac Operator on Collapsing Circle Bundles, Sém. Th. Spec.
Géom Inst. Fourier Grenoble 16 (1998), 33–42.
[2] B. Ammann, Spin-Strukturen und das Spektrum des Dirac-Operators, Ph.D. thesis,
University of Freiburg, Germany, 1998, Shaker-Verlag Aachen 1998, ISBN 3-
8265-4282-7.
[3] , The smallest Dirac eigenvalue in a spin-conformal class and cmc-
immersions, Comm. Anal. Geom. 17 (2009), 429–479.
[4] , A spin-conformal lower bound of the first positive Dirac eigenvalue, Diff.
Geom. Appl. 18 (2003), 21–32.
[5] , A variational problem in conformal spin geometry, Habilitationsschrift,
Universität Hamburg, 2003.
[6] B. Ammann and C. Bär, Dirac eigenvalues and total scalar curvature, J. Geom.
Phys. 33 (2000), 229–234.
[7] B. Ammann and E. Humbert, The first conformal Dirac eigenvalue on 2-
dimensional tori, J. Geom. Phys. 56 (2006), 623–642.
[8] B. Ammann, E. Humbert, and B. Morel, Mass endomorphism and spinorial Yam-
abe type problems, Comm. Anal. Geom. 14 (2006), 163–182.
[9] B. Ammann, A. D. Ionescu, and V. Nistor, Sobolev spaces on Lie manifolds and
regularity for polyhedral domains, Doc. Math. 11 (2006), 161–206.
[10] B. Ammann, R. Lauter, and V. Nistor, On the geometry of Riemannian manifolds
with a Lie structure at infinity, Int. J. Math. Math. Sci. (2004), 161–193.
[11] , Pseudodifferential operators on manifolds with a Lie structure at infinity,
Ann. of Math. 165 (2007), 717–747.
[12] C. Bär, The Dirac operator on space forms of positive curvature, J. Math. Soc.
Japan 48 (1996), 69–83.
[13] C. Bär, The Dirac operator on hyperbolic manifolds of finite volume, J. Differ.
Geom. 54 (2000), 439–488.
[14] H. Baum, Spin-Strukturen und Dirac-Operatoren über pseudoriemannschen Man-
nigfaltigkeiten, Teubner Verlag, 1981.
[15] A. L. Besse, Einstein manifolds, Ergebnisse der Mathematik und ihrer Grenzgebi-
ete, 3. Folge, no. 10, Springer-Verlag, 1987.
[16] T. P. Branson, Differential operators canonically associated to a conformal struc-
ture, Math. Scand. 57 (1985), 293–345.
[17] , Group representations arising from Lorentz conformal geometry, J. Funct.
Anal. 74 (1987), no. 2, 199–291.
[18] , Second order conformal covariants, Proc. Amer. Math. Soc. 126 (1998),
1031–1042.
[19] B. Colbois and A. El Soufi, Eigenvalues of the Laplacian acting on p-forms and
metric conformal deformations, Proc. of Am. Math. Soc. 134 (2006), 715–721.
[20] A. El Soufi and S. Ilias, Immersions minimales, première valeur propre du laplacien
et volume conforme, Math. Ann. 275 (1986), 257–267.
[21] H. D. Fegan, Conformally invariant first order differential operators., Quart. J.
Math. Oxford, II. series 27 (1976), 371–378.
[22] R. Gover and L. J. Peterson, Conformally invariant powers of the Laplacian,
Q-curvature, and tractor calculus, Comm. Math. Phys. 235 (2003), 339–378.
 The supremum of first eigenvalues 23

[23] E. Hebey and F. Robert, Coercivity and Struwe’s compactness for Paneitz type
operators with constant coefficients, Calc. Var. Partial Differential Equations 13
(2001), 491–517.
[24] N. Hitchin, Harmonic spinors, Adv. Math. 14 (1974), 1–55.
[25] P. Jammes, Extrema de valeurs propres dans une classe conforme, Sémin. Théor.
Spectr. Géom. 24 (2007), 23–42.
[26] I. Kolář, P. W. Michor, and J. Slovák, Natural operations in differential geometry,
Springer-Verlag, Berlin, 1993.
[27] N. Korevaar, Upper bounds for eigenvalues of conformal metrics, J. Differ. Geom.
37 (1993), 73–93.
[28] H.-B. Lawson and M.-L. Michelsohn, Spin Geometry, Princeton University Press,
1989.
[29] J. M. Lee and T. H. Parker. The Yamabe problem. Bull. Am. Math. Soc., New Ser.
17 (1987), 37–91.
[30] S. Mac Lane, Categories for the working mathematician, Graduate Texts in Math-
ematics, vol. 5, Springer-Verlag, New York, 1998.
[31] R. B. Melrose, The Atiyah-Patodi-Singer index theorem, Research Notes in Math-
ematics, vol. 4, A K Peters Ltd., Wellesley, MA, 1993.
[32] S. M. Paneitz, A quartic conformally covariant differential operator for arbitrary
pseudo-Riemannian manifolds, Preprint 1983, published in SIGMA 4 (2008).
[33] M. E. Taylor, M. E., Pseudodifferential operators, Princeton University Press,
1981.

Authors’ addresses:

Bernd Ammann
Facultät für Mathematik
Universität Regensburg
93040 Regensburg
Germany
[email protected]

Pierre Jammes
Laboratoire J.-A. Diendonné, Université Nice-Sophia Antipolis,
Parc Valrose, F-06108 Nice Cedex02, France
[email protected]
 2
K-Destabilizing test configurations with
smooth central fiber
claudio arezzo, alberto della vedova, and
gabriele la nave

Abstract
In this note we point out a simple application of a result by the authors in
[2]. We show how to construct many families of strictly K-unstable polarized
manifolds, destabilized by test configurations with smooth central fiber. The
effect of resolving singularities of the central fiber of a given test configuration
is studied, providing many new examples of manifolds which do not admit
Kähler constant scalar curvature metrics in some classes.

2.1 Introduction
In this note we want to speculate about the following Conjecture due to Tian-
Yau-Donaldson ([23], [24], [25], [7]):

Conjecture 2.1.1 A polarized manifold (M, A) admits a Kähler metric of

constant scalar curvature in the class c1 (A) if and only if it is K-polystable.

The notion of K-stability will be recalled below. For the moment it suffices to
say, loosely speaking, that a polarized manifold, or more generally a polarized
variety (V , A), is K-stable if and only if any special degeneration or test
configuration of (V , A) has an associated non positive weight, called Futaki
invariant and that this is zero only for the product configuration, i.e. the trivial
degeneration.
We do not even attempt to give a survey of results about Conjecture 2.1.1, but
as far as the results of this note are concerned, it is important to recall the reader
that Tian [24], Donaldson [7], Stoppa [22], using the results in [3] and [4], and
Mabuchi [17] have proved the sufficiency part of the Conjecture. Destabilizing
a polarizing manifold then implies non existence results of Kähler constant
scalar curvature metrics in the corresponding classes.

24
 K-Destabilizing test configurations 25

One of the main problems in this subject is that under a special degeneration
a smooth manifold often becomes very singular, in fact just a polarized scheme
in general. This makes all the analytic tool available at present very difficult to
use.
Hence one naturally asks which type of singularities must be introduced to
make the least effort to destabilize a smooth manifold without cscK metrics.
The aim of this note is to provide a large class of examples of special
degenerations with positive Futaki invariant and smooth limit. In fact we want to
provide a “machine” which associates to any special degeneration of a polarized
normal variety (V , A) with positive Futaki invariant a special degeneration for
a polarized manifold (M̃, Ã) with smooth central fiber and still positive Futaki
invariant.
To the best of our knowledge, before this work the only known examples of
special degeneration with non negative Futaki invariant and smooth central fiber
are the celebrated example of Mukai-Umemura’s Fano threefold ([18]) used
by Tian in [24] to exhibit the first examples of Fano manifolds with discrete
automorphism group and no Kähler-Einstein metrics (other Fano manifolds
with these properties have been then produced in [1]). In this case there exist
non trivial special degenerations with smooth limit and zero Futaki invariant
(hence violating the definition of K-stability). It then falls in the borderline
case, making this example extremely interesting and delicate. We stress that
our “machine” does not work in this borderline case, because a priori the Futaki
invariant of the new test configuration is certainly small (by [2]) but we cannot
control its sign.
To state our result more precisely we now recall the relevant definitions:

Definition 2.1.2 Let (V , A) be a n-dimensional polarized variety or scheme.

Given a one-parameter subgroup ρ : C∗ → Aut(V ) with a linearization on A

and denoted by w(V , A) the weight of the C∗ -action induced on top H 0 (V , A),
we have the following asymptotic expansions as k  0:

h0 (V , Ak ) = a0 k n + a1 k n−1 + O(k n−2 ) (2.1)

w(V , A ) = b0 k
k n+1
+ b1 k + O(k
n n−1
) (2.2)

The (normalized) Futaki invariant of the action is the rational number

b1 b0 a1
F (V , A, ρ) = − 2 .
a0 a0
Definition 2.1.3 A test configuration (X, L) for a polarized variety (V , A)
consists of a scheme X endowed with a C∗ -action that linearizes on a line
bundle L over X, and a flat C∗ -equivariant map f : X → C (where C has the
26 C. Arezzo, A. Della Vedova, and Gabriele La Nave

usual weight one C∗ -action) such that L|f −1 (0) is ample on f −1 (0) and we have
(f −1 (1), L|f −1 (1) )  (V , Ar ) for some r > 0.
When (V , A) has a C∗ -action ρ : C∗ → Aut(V ), a test configuration where
X = V × C and C∗ acts on X diagonally through ρ is called product configu-
ration.
Given a test configuration (X, L) we will denote by F (X, L) the Futaki
invariant of the C∗ -action induced on the central fiber (f −1 (0), L|f −1 (0) ).
If (X, L) is a product configuration as above, clearly we have F (X, L) =
F (V , A, ρ).
Definition 2.1.4 The polarized manifold (M, A) is K-stable if for each test
configuration for (M, A) the Futaki invariant of the induced action on the central
fiber (f −1 (0), L|f −1 (0) ) is less than or equal to zero, with equality if and only if
we have a product configuration.
A test configuration (X, L) is called destabilizing if the Futaki invariant of
the induced action on (f −1 (0), L|f −1 (0) ) is greater than zero.
Test configurations for an embedded variety V ⊂ PN endowed with the hyper-
plane polarization A can be constructed as follows. Given a one-parameter
subgroup ρ : C∗ → GL(N + 1), which induces an obvious diagonal C∗ -action
on PN × C, it clear that the subscheme
 
X = (z, t) ∈ PN × C | t = 0, (ρ(t −1 )z, t) ∈ V ⊂ PN × C,
is invariant and projects equivariantly on C. Thus considering the relatively
ample polarization L induced by the hyperplane bundle gives test configuration
for (V , A). On the other hand, given a test configuration (X, L) for a polarized
variety (V , A), the relative projective embedding given by Lr , with r sufficiently
large, realizes X as above (see details in [21]).
We can now describe our “machine”: consider a test configuration (X, L)
for a polarized normal variety (V , A) with F (X, L) > 0. Up to raise L to a
suitable power – which does not affect the Futaki invariant – we can suppose
being in the situation above with X ⊂ PN × C invariantly, and L induced by the
hyperplane bundle of PN . At this point we consider the central fiber X0 ⊂ PN ,
which is invariant with respect to ρ, and we apply the (equivariant) resolution of
singularities [14, Corollary 3.22 and Proposition 3.9.1]. Thus there is a smooth
manifold P̃ acted on by C∗ and an equivariant map
β : P̃ → PN
which factorizes through a sequence of blow-ups, such that the strict transform
X̃0 of X0 is invariant and smooth. The key observation is that the strict transform
X̃1 of the fiber X1 ⊂ X degenerate to X0 under the given C∗ action on P̃ , thus it
 K-Destabilizing test configurations 27

must be smooth. This gives an invariant family X̃ ⊂ P × C and an equivariant

birational morphism

π : X̃ → X.

Some comments are in order:

1 all the fibers of X̃ are smooth, but π is never a resolution of singularities of

X (except the trivial case when the central fiber of X was already smooth)
since it fails to be an isomorphism on the smooth locus of X;
2 L̃ = π ∗ L is not a relatively ample line bundle any more, but just a big and nef
one. It is not then even clear what it means to compute its Futaki invariant;
3 the fiber over the generic point of C of the new (big and nef) test configuration
(X̃, L̃) is different from V ;
4 the family X̃ is not unique since the resolution β it is not.

The issue raised at point (2) was addressed in [2] and it was proved that the
following natural (topological) definition makes the Futaki invariant a continu-
ous function around big and nef points in the Kähler cone. We will give simple
self-contained proofs in the cases of smooth manifolds and varieties with just
normal singularities in Section 2.

Definition 2.1.5 Let V be a projective variety or scheme endowed with a C∗ -

action and let B be a big and nef line bundle on V . Choosing a linearization of
 V j
the action on B gives a C∗ -representation on dim j =0 H (V , B )
k (−1)j
(here the
E −1 denotes the dual of E). We set w(V , B k ) = tr Ak , where Ak is the generator
of that representation. As k → +∞ we have the following expansion

w(V , B k )
= F0 k + F1 + O(k −1 ),
χ (V , B k )

and we define

F (V , B) = F1

to be the Donaldson–Futaki invariant of the chosen action on (V , B)

The existence of the expansion involved in definition above follows from the
standard fact that χ (V , B k ) is a polynomial of degree dim V , whose proof (see
for example [11]) can be easily adapted to show that w(V , B k ) is a polynomial
of degree at most dim V + 1.
28 C. Arezzo, A. Della Vedova, and Gabriele La Nave

The key technical Theorem proved in [2] is then the following:

Theorem 2.1.6 Let B, A be linearized line bunldes on a scheme V acted on

by C∗ . Suppose that B is big and nef and A ample. We have

1
F (V , B r ⊗ A) = F (V , B) + O , as r → ∞.
r
Having established a good continuity property of the Futaki invariant up to these
boundary point, we need to address the question of the effect of a resolution
of singularities of the central fiber. This is a particular case of the following
non trivial extension of previous analysis by Ross and Thomas [21] which was
proved in [2] where the general case of birational morphisms has been studied:

Theorem 2.1.7 Given a test configuration f : (X, L) → C as above, let f  :

(X , L ) → C be another flat equivariant family with X normal and let β :
(X , L ) → (X, L) be a C∗ -equivariant birational morphism such that f  =
f ◦ β and L = β ∗ L. Then we have

F (X , L ) ≥ F (X, L),

with strict inequality if and only if the support of β∗ (OX )/OX has codimension
one.

The proof of these results uses some heavy algebraic machinery, yet their proof
when (V , A) or the central fiber of (X, L) have only normal singularities (a
case largely studied) is quite simple and we give it in Section 2.
The Corollary of Theorem 2.1.6 and Theorem 2.1.7 we want to point out in
this note is then the following:

Theorem 2.1.8 Let (X, L) be a test configuration for the polarized normal
variety (V , A) with positive Futaki invariant. Let moreover (X̃, L̃) be a (big
and nef) test configuration obtained from (X, L) as above and let (M̃, B̃) be
the smooth (big and nef) fiber over the point 1 ∈ C. Let R be any relatively
ample line bundle over X̃.
Then (X̃, L̃r ⊗ R) is a test configuration for (M̃, B̃ r ⊗ R|M̃ ) with following
properties:

1 smooth central fiber;

2 positive Futaki invariant for r sufficiently large.

In particular M̃ does not admit a constant scalar curvature Kähler metric in

any class of the form c1 (B̃ r ⊗ R|M̃ ), with r large enough.
 K-Destabilizing test configurations 29

While this Theorem clearly follows from Theorems 2.1.6 and Theorem 2.1.7,
but for the specific case of central fiber with normal singularities it follows
from the much simpler Proposition 2.2.1 and Theorem 2.2.3.
The range of applicability of the above theorem is very large. We go through
the steps of the resolution of singularities in an explicit example by Ding-
Tian [6] of a complex orbifold of dimension 2. In this simple example explicit
calculations are easy to perform, yet we point out that the final example is
somehow trivial since it ends on a product test configuration. On the other
hand abundance of similar examples even in dimension 2 can be obtained by
the reader as an exercise using the results of Jeffres [12] and Nagakawa [19],
in which cases we loose an explicit description of the resulting destabilized
manifold, but we get new nontrivial examples. In fact in higher dimensions one
can use the approach described in this note to test also the Arezzo-Pacard blow
up theorems [3] [4], when the resolution of singularities requires a blow up of
a scheme of positive dimension.

2.2 The case of normal singularities

In this section we give simple proofs of the continuity of the Futaki invariant
at boundary points for smooth manifolds or varieties with normal singularities.
More general results of this type have been proved in [2] but we want to stress
that under these assumptions proofs become much easier.
The fundamental continuity property we will need, and proved in Corollary
2.1.6, can be stated in the following form for smooth bases:

Proposition 2.2.1 Let A, L be respectively an ample and a big and nef line
bundle on a smooth projective manifold M. For every C∗ -action on M that
linearizes to A and L, as r → +∞ we have
1
F (M, Lr ⊗ A) = F (M, L) + O .
r
Proof The result is a simple application of the equivariant Riemann-Roch
Theorem. We present here the details of the calculations involved, since we
could not find precise references for them.
Fix an hermitian metrics on A that is invariant with respect to the action of
S 1 ⊂ C∗ and suppose that the curvature ω is a Kähler metric. Since L is nef,
for each r > 0 we can choose an invariant metric on L whose curvature ηr
satisfy rηr + ω > 0. In other words rηr + ω is a Kähler form which coincides
with the curvature of the induced hermitian metric on the line bundle Lr ⊗ A.
30 C. Arezzo, A. Della Vedova, and Gabriele La Nave

Setting n = dim(M), by Riemann–Roch for each r > 0 we have:


(ηr + 1r ω)n
χ (M, L ⊗ A ) = (rk)
rk k n
M n!

(ηr + 1r ω)n−1 ∧ Ric(ηr + 1r ω)
+ (rk)n−1 + O(k n−2 ).
M 2(n − 1)!

Now let v be the holomorphic vector field on M generating the given C∗ -

action. Let f and gr be smooth S 1 -invariant functions on M such that
iv ω + ∂f ¯ r = 0. They exist since the given C∗ -action on
¯ = 0 and iv ηr + ∂g
M lifts to A and L ⊗ A (see [7, pag. 294]). Applying, in a similar fashion,
r

the equivariant Riemann-Roch theorem, the weight of the induced action on

n k (−1)j
j =0 det H (M, L ⊗ A )
j rk
is calculated as:

1 (ηr + 1r ω)n
w(M, L ⊗ A ) = (rk)
rk k n+1
gr + f
M r n!

1 (ηr + 1r ω)n−1 ∧ Ric(ηr + 1r ω)
+ (rk)n gr + f
M r 2(n − 1)!
+ O(k n−1
).

As r → +∞ we have expansions:

(ηr + 1r ω)n c1 (Lr ⊗ A)n c1 (L)n 1
= n
= +O
M n! n! r n! r

(ηr + 1
ω)n−1 ∧ Ric(ηr + 1r ω) c1 (Lr ⊗ A)n−1 c1 (M) 1
r
= +O
M 2(n − 1)! 2(n − 1)! r n−1 r
c1 (L)n−1 c1 (M) 1
= +O
2(n − 1)! r

1 (ηr + 1r ω)n cT (Lr ⊗ A)n+1
gr + f = 1
M r n! (n + 1)! r n+1
c1T (L)n+1 1
= +O
(n + 1)! r

1 (ηr + 1r ω)n−1 ∧ Ric(ηr + 1r ω) cT (Lr ⊗ A)n c1T (M)
gr + f = 1
M r 2(n − 1)! 2 n! r n
c1T (L)n c1T (M) 1
= +O ,
2 n! r
 K-Destabilizing test configurations 31

where c1T denotes the equivariant first Chern class. Thus we have:
c1T (L)n c1T (M) c1 (L)n c1T (L)n+1 c1 (L)n−1 c1 (M)
2n!
· n!
− (n+1)!
· 2(n−1)! 1
F (M, L ⊗ A) =
r
 c1 (L)n 2 +O .
r
n!

and the thesis follows applying again the (equivariant) Riemann-Roch theorem
to L. 

The following has essentially been proved by Paul-Tian ([20]); here we give a
simple self-contained proof.

Proposition 2.2.2 Let (V , L) be a polarized normal projective variety endowed

with a C∗ -action. Let π : Ṽ → V be an equivariant resolution of singularities.
We have

F (V , L) = F (Ṽ , π ∗ L).

Proof By Zariski’s Main theorem we have an equivariant isomorphism

π∗ OṼ  OV , moreover normality of V implies dim Supp R q π∗ (OṼ ) ≤ n − 1
for all q > 0, where n + 1 = dim V = dim Ṽ . In fact, since π is an isomor-
phisms outside the singularities of V , Supp R q π∗ (OṼ ) is contained in the
singular locus of V , which by normality has dimension less than or equal to
n − 1.
Thus the projection formula yields:

χ (Ṽ , π ∗ Lk ) = χ (V , Rπ∗ (OṼ ) ⊗ Lk ) = χ (V , Lk ) + O(k n−2 ),

and by the same token:

w(Ṽ , π ∗ Lk ) = w(V , Lk ) + O(k n−1 ).

The thesis follows by definition 2.1.5. 

Combining Propositions 2.2.1 and 2.2.2 we get the following result, which
explains the behaviour for normal varieties of the Futaki invariant under Hiron-
aka’s resolution of singularities process:

Theorem 2.2.3 Let (V , L) be a polarized normal variety endowed with a

C∗ -action. Let π : Ṽ → V be an equivariant resolution of singularities with
exceptional divisor E. Then we have
  1
F Ṽ , π ∗ Lr ⊗ O (−E) = F (V , L) + O as r → +∞.
r
32 C. Arezzo, A. Della Vedova, and Gabriele La Nave

In particular, if F (V , L) = 0, then Ṽ admits no cscK metrics in the classes

rπ ∗ c1 (L) − E for r  0.

Proof Since π is a sequence of blow-ups along smooth centers, π ∗ L is big and

nef, moreover there exists r0 > 0 such that π ∗ Lr0 ⊗ O(−E) is ample. Thus, as
r → +∞ we obtain:
   
F Ṽ , π ∗ Lr ⊗ O (−E) = F Ṽ , π ∗ Lr−r0 ⊗ π ∗ Lr0 ⊗ O (−E)
  1
= F Ṽ , π ∗ Lr−r0 + O
r
1
= F (V , L) + O ,
r

where the second and third equalities follow from Propositions 2.2.1 and 2.2.2
respectively. 

2.3 Proof of Theorem 2.1.8 and examples

Proof of Theorem 2.1.8 On each fiber of X̃ → C the restrictions of L̃r and R
are nef and ample line bundles respectively, thus L̃r ⊗ R is relatively ample and
gives, together with X, a test configuration for (M̃, B̃ r ⊗ R), where B̃ = L̃|M̃ .
All the fibers of that test configuration are smooth by construction thus
point (1) is proved. Moreover by Theorem 2.1.6 (or Proposition 2.2.1 since the
central fiber of X̃ is smooth) we have the asymptotic expansion as r → +∞

1
F (X̃, L̃r ⊗ R) = F (X̃, L̃) + O . (2.3)
r

On the other hand, the morphism π : X̃ → X is equivariant and birational, and

L̃ = π ∗ L, so by Theorem 2.1.7 we get

F (X̃, L̃) ≥ F (X, L) > 0. (2.4)

Thus point (2) follows readily from (2.3) and (2.4). 

Let us now see an explicit appearance of the phenomenon described in Theorem

2.1.8. We construct K-destabilizing degenerations with smooth central fiber of
a smooth surface obtained by resolution of singularities of a mild singular
unstable cubic surface found by Ding-Tian ([6]).
Let us the look at Xf ⊂ P3 be the zero locus of f = z0 z12 + z2 z3 (z2 − z3 ).
 K-Destabilizing test configurations 33

Let us collect some elementary facts that the reader can easily verify:
1 Xf is singular only at p0 = (1 : 0 : 0 : 0).
2 Having set (x, y, z) affine co-ordinates centered on (1 : 0 : 0 : 0), let
C  = {((x, y, z), (l0 : l1 : l2 )) ∈ C3 × P2 |
xl1 − yl0 = xl2 − zl0 = yl2 − zl1 = 0}
be the blow-up of C3 at the origin with exceptional divisor E  P2 and Xf
the proper transform of Xf , then Xf is singular at points
(0 : 1 : 0), (0 : 0 : 1), (0 : 1 : 1) ∈ E.
3 Having set X̃f the proper transform of Xf under the blow-up C̃ of C  at
points (0 : 1 : 0), (0 : 0 : 1), (0 : 1 : 1) ∈ E and E1 the exceptional divisor
over (0 : 1 : 0), then X̃f is smooth around E1 . Analogously, if E2 is the
exceptional divisor over (0 : 0 : 1) and E3 the one over (0 : 1 : 1), then X̃f
is smooth around E2 and E3 too. For future reference let E0 be the proper
transform of E under the second blow up.
4 X̃f is smooth.
5 Consider now the C∗ -action on P3 defined by
t · (z0 , z1 , z2 , z3 ) = (t α0 z0 , t α1 z1 , t α2 z2 , t α3 z3 )
with αj ∈ Z and α0 + · · · + α3 = 0, thus
(t · f )(z) = t −(α0 +2α1 ) z0 z12 + t −(2α2 +α3 ) z22 z3 − t −(α2 +2α3 ) z2 z32 ,
and f is semi-invariant if and only if
α0 + 2α1 = 2α2 + α3 = α2 + 2α3 ,
hence
(α0 , . . . , α3 ) = (−7β, 5β, β, β),
with β ∈ Z, and f has weight −3β.
6 The monomials of degree three in the variables z0 , . . . , z3 are a basis of
semi-invariants for the fixed C∗ -action on C[z0 , . . . , z3 ]3 . In particular fix
β = −1, then the subspace spanned by monomials with weight greater or
equal to 4 is
V = span{z13 , z12 z2 , z12 z3 , z1 z22 , z1 z2 z3 , z1 z32 }
Thus the hypersurfaces that degenerate to Xf under the fixed C∗ -action (with
β = −1) are of the form
Xg = {f + g = 0},
where g ∈ V .
34 C. Arezzo, A. Della Vedova, and Gabriele La Nave

7 The general Xg has a rational double point at p0 = (1 : 0 : 0 : 0) and is

smooth elsewhere.
8 X̃f is the flat limit of X̃g in P̃ , the iterated blow-up of P3 as described
(locally) above.

Thus X̃f is the central fiber of the special degeneration of X̃g given by the chosen
C∗ -action. Denoted byπ4 : P̃ →  P the blow-up map, let Ar be the restriction
3
∗
of π OP3 (r) ⊗ O − =0 Ej to X̃f . By Theorem 2.2.3, F (X̃f , Ar ) has the
same sign of F (Xf , OXf (1)) when r is large enough. But thanks to [6] (see also
[16]), with our sign convention, (Xf , OXf (1)) has positive Futaki invariant, then
  
X̃g polarized with the restriction of π ∗ OP3 (r) ⊗ O − 4=0 Ej is K-unstable.

References
[1] C. Arezzo, A. Della Vedova, On the K-stability of complete intersections in polar-
ized manifolds, arXiv:0810.1473.
[2] C. Arezzo, A. Della Vedova and G. La Nave, Singularities and K-semistability,
arXiv:0906.2475.
[3] C. Arezzo and F. Pacard, Blowing up and desingularizing Kähler orbifolds with
constant scalar curvature, Acta Mathematica 196, no 2, (2006) 179-228.
[4] C. Arezzo and F. Pacard, Blowing up Kähler manifolds with constant scalar
curvature II, Annals of Math. 170 no. 2, (2009) 685-738
[5] D. Eisenbud, J. Harris, The Geometry of schemes. GTM 197. Springer, 2000.
[6] W. Y. Ding and G. Tian, Kähler-Einstein metrics and the generalized Futaki
invariant. Invent. Math. 110 (1992), no. 2, 315–335.
[7] S. K. Donaldson, Scalar curvature and stability of toric varieties. J. Differential
Geom. 62 (2002), no.2, 289–349.
[8] S. K. Donaldson. Lower bounds on the Calabi functional, J. Differential Geom.,
70 (2005), no.3, 453–472.
[9] J. Fine and J. Ross, A note on positivity of the CM line bundle. Int. Math. Res.
Not., 2006.
[10] A. Futaki and T. Mabuchi, Bilinear forms and extremal Kähler vector fields asso-
ciated with Kähler classes. Math. Ann. 301 (1995), n.2, 199–210
[11] R. Hartshorne, Algebraic Geometry, Springer, 1977.
[12] T. Jeffres, Singular set of some Kähler orbifolds. Trans. Amer. Math. Soc. 349
(1997), no. 5, 1961–1971.
[13] G. Kempf, F. Knudsen, D. Mumford and B. Saint-Donat, Toroidal Embeddings I
Lecture Notes in Mathematics, 339. Springer, 1973.
[14] J. Kollár, Lectures on resolution of singularities. Annals of Mathematics Studies,
166. Princeton University Press, Princeton, NJ, 2007.
[15] R. Lazarsfeld. Positivity in Algebraic Geometry I. Springer, 2004.
[16] Z. Lu, On the Futaki invariants of complete intersections. Duke Math. J. 100
(1999), no.2, 359–372.
 K-Destabilizing test configurations 35

[17] T. Mabuchi, K-stability of constant scalar curvature polarization,

arXiv:0812.4093.
[18] S. Mukai and H. Umemura, Minimal rational threefolds. Algebraic geometry
(Tokyo/Kyoto, 1982), 490–518, Lecture Notes in Math., 1016, Springer, Berlin,
1983.
[19] Y. Nakagawa, Bando-Calabi-Futaki characters of Kähler orbifolds. Math. Ann.
314 (1999), no. 2, 369–380.
[20] S. Paul and G. Tian, CM Stability and the Generalized Futaki Invariant I. math.DG/
0605278
[21] J. Ross and R. P. Thomas, An obstruction to the existence of constant scalar
curvature Kähler metrics. Jour. Diff. Geom. 72, 429–466, 2006.
[22] J. Stoppa, J. Stoppa, K-stability of constant scalar curvature Kaehler manifolds,
Advances in Mathematics 221 no. 4 (2009), 1397–1408.
[23] G. Tian, Recent progress on Kähler-Einstein metrics, in Geometry and physics
(Aarhus, 1995), 149-155, Lecture Notes in Pure and Appl. Math., 184, Dekker,
New York, 1997.
[24] G. Tian, Kähler-Einstein metrics with positive scalar curvature. Invent. Math. 130
(1997), no. 1, 1–37.
[25] G. Tian, Extremal metrics and geometric stability. Houston Math. J. 28 (2002),
no. 1, 411–432.

Authors’ addresses:

Claudio Arezzo
Abdus Salam International Center for Theoretical Physics,
Strada Costiera 11,
Trieste (Italy) and Dipartimento di Matematica,
Università di Parma,
Parco Area delle Scienze 53/A,
Parma, Italy
[email protected]

Alberto Della Vedova

Fine Hall,
Princeton University,
Princeton, NJ 08544
and
Dipartimento di Matematica,
Università di Parma,
Parco Area delle Scienze 53/A,
Parma, Italy
[email protected]
36 C. Arezzo, A. Della Vedova, and Gabriele La Nave

Gabriele La Nave
Department of Mathematics
Yeshiva University
500 West 185 Street
New York, NY
USA
[email protected]
 3
Explicit constructions of Ricci solitons
paul baird

Abstract
We describe methods for constructing explicit examples of Ricci solitons. An
improved version of an anzatz of the author and L. Danielo for 3-dimensional
solitons is given which we extend to higher dimensions. The soliton structure
on the 4-dimensional geometry Nil4 is analysed in detail, in particular its
uniqueness is established and its relation to the natural Riemannian projection
Nil4 → Nil3 is discussed.

3.1 Introduction
The Ricci flow is the evolution equation:

∂g
= −2Ricci (g) (3.1)
∂t
for a time-dependent Riemannian metric g = g(t) defined on an n-dimensional
manifold M n , subject to some initial condition g(0) = g0 . This equation has
received a considerable amount of attention recently due to the work of G.
Perelman [14] which has advanced Hamilton’s program for resolving the
geometrization conjecture. An essential part of this story are Ricci solitons,
which correspond to fixed points, up to scaling and diffeomorphism. Specifi-
cally, if g(t) = c(t)ψt∗ (g0 ) is a solution of (3.1) on some time interval [0, δ),
where c(t) is a family of positive scalars such that c(0) = 1 and ψt is a family
of diffeomorphisms satisfying ψ0 = id, then on calculating g  (0) and applying
(3.1) we obtain the equation:

2Ricci (g) + LE g + 2Ag = 0 , (3.2)

37
38 P. Baird

where we now write g = g0 , and where 2A = c (0) and E is the vector field
(called the soliton flow) determined at each x ∈ M by E(x) = dtd ψt (x)|t=0 .
Conversely, on a complete manifold, any solution g0 of (3.2) determines
a solution of (3.1) of the form g(t) = c(t)ψt∗ (g0 ) for some small time t
(see [5]).
Equation (3.2) is known as the soliton equation and solutions play a fun-
damental role in the study of the Ricci flow: they occur as rescaled limits at
singularity formation and as asymptotic limits of immortal solutions, that is
solutions that exist for all future time [13]. Both of these limits are interpreted
in terms of Cheeger-Gromov-Hamilton pointed convergence of Ricci flows [9].
In the case when E = grad f is the gradient of a function, then the soliton is
said to be of gradient type, in which case (3.2) becomes

Ricci (g) + ∇ 2 f + Ag = 0 , (3.3)

since LE g = 2∇ 2 f where ∇ 2 f denotes the Hessian of f . A soliton defined by

(3.2) is called shrinking, steady or expanding according as the constant A is
negative, zero or positive, respectively.
At singularity formation, it is only gradient solitons of a specific type which
occur, moreover, a gradient soliton is also a critical metric for Perelman’s
entropy functional [14]. Non-gradient solitons are important in the study of the
long-term behaviour of solutions and it is only recently that the first examples
were found [2, 13]. The existence of such solitons also has important conse-
quences for the stability of the Ricci flow about non-Ricci flat metrics [8]; see
also [7, 15] for the Ricci-flat case.
Clearly the soliton flow E is only defined up to addition of a Killing vector
field and if it is itself Killing, then equation (3.2) becomes the equation for
an Einstein metric. These solutions are referred to as the trivial Ricci solitons.
In dimension 3, any compact soliton is trivial [10]. The greater flexibility in
equation (3.2) makes the problem of finding non-trivial Ricci solitons both
rich and challenging. Even in dimension 2, the complete solitons are not yet
classified; see [5] Chapter 1, for a discussion.
One way to find explicit solutions to (3.2) is to suppose that the metric is
sufficiently symmetric that the equations reduce to an ODE. The best-known
example of such a construction is given by the Bryant soliton, which, in dimen-
sion two corresponds to the Hamilton cigar. In this case one considers a warped
product metric of the form g = dr 2 + F (r)2 gS n , where gS n is the standard metric
on the n-sphere. One looks for a gradient soliton with soliton flow E = grad f
with f depending only on the parameter r. The system of equations to solve
then becomes a dynamical system in R2 . An orbit emanating from one of the
 Explicit constructions of Ricci solitons 39

fixed points gives a complete metric of positive sectional curvature [4] (see Sec-
tion 3.4). This procedure was generalized by T. Ivey to construct solitons from
doubly warped product metrics of the type dr 2 + a(r)2 g1 + b(r)2 g2 , where
g1 and g2 are metrics on a sphere and an Einstein manifold, resp. [11]. In
Section 3.2, we make a further generalization which leads to a dynamical sys-
tem. Here we no longer insist that the soliton be of gradient type and neither
do we require that the soliton flow depend on a single parameter. The proce-
dure picks up solitons of Sol-type, as well as a curious example on a surface.
Although the soliton metric is neither complete nor compact, it factors and then
extends by the addition of two points to a non-standard (complete) metric on
the 2-sphere (Example 3.2.6).
In [2], the author and L. Danielo devised a procedure for constructing
solitons in dimension 3, by reducing the number of variables to two, so that
(3.2) becomes a system of equations on a Riemannian surface. It is important
to note that the method does not consist simply of supposing quantities depend
on two independent variables, rather, having solved the system of equations
on the surface one is then required to solve an exterior differential equation in
dimension 3 in order to obtain the 3-dimensional soliton metric. In particular,
the metric itself and the soliton flow may depend on all three coordinates. In
Section 3.3, we give an improved version of this ansatz and describe how
the soliton structure on Nil3 arises. Then in Section 3.4, we generalize the
construction to more general dimensions.
On Rn there are the well-known Gaussian solitons with flow E =
grad(− A2 |x − a|2 ), where a ∈ Rn is some arbitrary point and A is an arbi-
trary constant. This shows that a soliton structure may not be unique. However,
it is not known whether there are other examples of Riemannian manifolds
which support more than one soliton structure. One way to test this for a given
manifold, is to solve the equation (3.2) directly. This requires sufficient sym-
metry of the metric in order that the system of equations is managable. In [2],
this was done for the 3-dimensional geometries and uniqueness was established
in all cases except R3 , where one readily sees that the only examples are the
Gaussian solitons. In Section 3.5, we do this for the 4-dimensional geometry
Nil4 , to obtain the soliton structure whose existence was demonstrated by J.
Lauret [12] by Lie group methods and is described more explicitly by J. Lott
[13]. We further show uniqueness of this structure and discuss its relation to
the one on Nil3 : the geometry Nil4 naturally admits a harmonic Riemannian
fibration to Nil3 , which allows us to relate their respective soliton structures.
The author thanks the Agence Nationale de Recherche project: Flots et
opérateurs géométriques no: ANR-07-BLAN-0251-01 for financial support
during the preparation of this paper.
40 P. Baird

3.2 Solitons from a dynamical system

Let us consider a 3-dimensional metric g = gij dx i dx j , which, in coordinates
(x 1 , x 2 , x 3 ) is of diagonal form:
⎛ ⎞
1 0 0
(gij ) = ⎝ 0 a 2 0 ⎠ (3.4)
0 0 b2

where the functions a, b both depend on x 1 only. As we shall see, the vector
field E defining the soliton flow may also depend on the variable x 3 . Then the
components of the Ricci curvature are given by the formula:

Rj k = ∂l jl k − ∂j lk
l
+ lm
l
jmk − jl m lk
m
,

and we find that the Ricci curvature is also diagonal with components:

R11 = −(u + v  + u2 + v 2 ), R22 = −a 2 (v  + uv + v 2 ), R33 = −b2 (u + u2 + uv).

In order to obtain a dynamical system, we suppose that the soliton flow E in

(3.2) has the form:

E = X + fU (3.5)

where U = b1 ∂3 and f is some function. We suppose further that X = grad β,

where β = β(x 1 , x 2 ). This is not the same as supposing that E is of gradient
type, but rather that it is of gradient type modulo a component in the direction
of the vector field U . The reason that we make this decomposition is that, by an
application of the Poincaré Lemma, it is possible to eliminate the function f
from (3.2), so reducing the number of unknown parameters by one. A further
useful consequence is to essentially fix a gauge for E, since in general, the
soliton flow is defined only up to addition of a Killing vector field. We find that
orbits of our dynamical system are independent of any freedom that remains in
the choice of E. The following theorem is proved in [1]

Theorem 3.2.1 Let A and k be constants and consider the 3-dimensional

autonomous dynamical system:
⎧ 
⎨ u = uρ − k
v  = vρ + A (3.6)
⎩ 
ρ = u2 + v 2 − A .
Then each orbit t → (u(t), v(t), ρ(t)) determines a Ricci soliton metric of the
form g = (dx 1 )2 + a(t)2 (dx 2 )2 + b(t)2 (dx 3 )2 via the correspondence: t = x 1 ,
u = b /b, v = a  /a. The soliton is shrinking, steady or expanding, according
 Explicit constructions of Ricci solitons 41

as A < 0, = 0, > 0, respectively. Up to addition of a Killing vector field, the

soliton flow is given by:
∂ ∂
E = (ρ + u + v) − (k + A)x 3 3 ; (3.7)
∂x 1 ∂x
it is of gradient type if and only if either k + A = 0, or b = 0 (u = 0) .

A first observation is that the constant A in (3.6) can be normalized by

permitting a homothetic change of the metric g. More precisely, set s = ct (c
constant), so that
ds 2
g= + a(t)2 (dx 2 )2 + b(t)2 (dx 3 )2
c2
1  
= 2 ds 2 + a (s)2 (dx 2 )2 + 
b(s)2 (dx 3 )2
c
where we have written  a (s) = ca(s/c),  b(s) = cb(s/c). If we define 
u(s) =
 
b (s)/b(s) = u(t)/c, 
v (s) = a  (s)/
a (s) = v(t)/c. Then
1  1 k
u (s) =
 2
u (t) = 2 (uρ − k) = − 2
uρ
c c c
(s) = ρ(s/c)/c. Similarly for the other derivatives and
where we have defined ρ
the dynamical system (3.6) now becomes:
⎧ 
⎨u (s) =   − ck2
uρ

v (s) =  + cA2
vρ (3.8)
⎩ 
 (s) = 
ρ u2 + 
v 2 − cA2
By choosing c appropriately, we can now normalise the constant A to be −2, 0
or 2.
We now show how a number of solitons metrics, some trivial, some non-
trivial, arise as orbits of (3.6).

Example 3.2.2 (The geometry Sol) For A > 0, the equilibrium points are
given by
√ √ √ !
±k A ∓A A k 2 + A2
√ ,√ ,± √ .
k 2 + A2 k 2 + A2 A
This gives the metric
⎛ ⎞
1 0
#
0
⎜ A ⎟
⎜0 e
2A
k 2 +A2
t
0 ⎟
⎝ # ⎠
−2k A
t
0 0 e k 2 +A2
42 P. Baird

But after normalisation of the constant A, we may suppose that

%
A
A 2 = 1.
k + A2
Now write λ = k/A to give the form:
⎛ ⎞
1 0 0
⎝ 0 e2t 0 ⎠
0 0 e−2λt

where λ is an arbitrary parameter. In the case when λ = 1 this gives the metric
for the geometry Sol, in particular, Sol is one of a 1-parameter family of soliton
metrics.

Example 3.2.3 (Constant curvature examples) The sphere S 3 arises by setting

A = −2, k = 2, a = sin t and b = cos t. Then u = − tan t, v = cot t and the
corresponding orbit of (3.6) is given by setting ρ = sin t1cos t − 2 cot t.
In a similar fashion, hyperbolic 3-space occurs by setting A = 2, k = −2,
a = sinh t and b = cosh t. Then u = tanh t, v = coth t and the corresponding
orbit occurs by setting ρ = − sinh t1cosh t − 2 tanh t.
Note that in both cases, the orbits are not defined for all time. Also, in both
cases, with reference to Theorem 3.2.1, we have k + A = 0.

Example 3.2.4 (Product examples) Examples of the form  × R occur when

b ≡ 1. Then u ≡ 0 and we obtain the 2-dimensional dynamical system:
 
v = vρ + A
ρ  = v2 − A .

But this is precisely the system that occurs in the construction of a 2-

dimensional Ricci soliton with metric of the form: h = (dx 1 )2 + a(x 1 )2 (dx 2 )2 .
For example, Hamilton’s 2-dimensional cigar is given by the metric: h =
(dx 1 )2 + tanh2 t(dx 2 )2 . Then it is easily seen that the above equations are
satisfied with A = 0. This example is a complete steady gradient soliton; for
more details, see for example [6].

Example 3.2.5 (Warped product examples) We consider the special case when
a ≡ b, which can be considered as a warped product of the real line with
standard Euclidean 2-space. Now our system can be viewed as the sub-system
of the dynamical system (3.6) given by the orbits passing through the axis
u = v = 0 when A + k = 0. It is easily seen that the plane u = v is an invariant
subspace, so the orbits remain in this plane. This then leads to the 2-dimensional
 Explicit constructions of Ricci solitons 43

dynamical system:

u = uρ + A
ρ  = 2u2 − A .
By eliminating ρ, we see that u is given as a solution to the second order ODE:
u u − (u )2 + Au + Au2 − 2u4 = 0.
A particular solution when A = 0 is given by u = ± √12 t −1 , which yields
√
the metric g = dt 2 + t ± 2 ((dx 2 )2 + (dx 3 )2 ). This singular metric of negative
scalar curvature was also noticed in [2]. It corresponds to a steady gradient 3-
dimensional soliton. If we don’t insist on completeness, this example provides
an answer to Problem 1.88 of [5]; by the expressions for the components of the
Ricci curvature, we see that the Ricci tensor is positive whatever sign is taken
in the expression for u, as is required – cf. the discussion in [5].
Example 3.2.6 (Non-standard structure over S 2 ) Let us make the orthogonal
substitution:
Au + kv −ku + Av
ξ=√ , η= √
k +A
2 2 k 2 + A2
which puts (3.6) into the form:
⎧ 
⎨ ξ = ξρ √
η = ηρ + k 2 + A2 (3.9)
⎩ 
ρ = ξ 2 + η2 − A .
We study the orbit with initial condition (ξ (0), η(0), ρ(0)) = (0, 0, 0), so that
ξ (t) = 0 for all t, and we now have the reduced system:
  √
η = ηρ + k 2 + A2
(3.10)
ρ  = η2 − A .
% !
√ k 2 + A2
If A > 0, the fixed points are given by ± A, ∓ , otherwise there
A
are none unless A = k = 0 in which case the whole axis η = 0 consists of fixed
points. We study the case A = 0, k = 0; for the other cases, see [1].
The dynamical system (3.10) now takes the form
 
η = ηρ + k
(3.11)
ρ  = η2 .
On replacing η with −η, it is no loss of generality to suppose that k > 0. We
consider the orbit with initial condition (η(0), ρ(0)) = (0, 0). Then there exists
a T > 0 such that η(t), ρ(t) → ∞ as t → T − . In [1] it is shown that this orbit
44 P. Baird

determines a steady soliton metric g of non-constant curvature on the surface

(0, 2T ) × R. The soliton is of non-gradient type and its flow is given by the
vector field
∂ ∂
E = (η − ρ) − kx 3 3 ,
∂r ∂x
where r = T − t. In fact limr→0+ (η(r) − ρ(r)) = limr→2T − (η(r) − ρ(r)) = 0.
The metric g factors to a metric on the cylinder (0, 2T ) × S 1 and then by the
cylinder to ball rule, extends to a C 2 metric on the sphere S 2 .

3.3 Reduction of the equations to a 2-dimensional system

In [2], an ansatz is given for constructing a soliton metric on a 3-manifold M 3
from data on a surface N 2 . The basis of this ansatz is the relation between objects
on N 2 and M 3 given that there exists a semi-conformal mapping ϕ : M 3 → N 2 .
In this section we give an improved version of this ansatz which highlights
how the soliton equation on M 3 corresponds to a “deformed” soliton equation
on N 2 .
A smooth map ϕ : (M m , g) → (N n , h) between manifolds M, N with
respective Riemannian metrics g, h is said to be semi-conformal if, for each
x ∈ M where dϕx = 0, the restriction dϕx |Hx : Hx → Tϕ(x) N is conformal and
surjective, where Hx = (ker dϕx )⊥ . Thus there exists a number λ = λ(x) > 0
such that ϕ ∗ h(X, Y ) = λ(x)2 g(X, Y ) for all X, Y ∈ Hx . Setting λ = 0 at points
x ∈ M where dϕx = 0, we obtain a continuous function λ : M → R (≥ 0)
called the dilation of ϕ; it has the property that λ2 = n1 ||dϕ||2 is smooth,
where ||dϕ|| denotes the Hilbert-Schmidt norm of the derivative at each
point. The fundamental equation of a semi-conformal submersion relates
the tension field, the dilation and the mean curvature μ of fibres by the
formula [3]:
τ (ϕ) = −(n − 2)dϕ(grad ln λ) − (m − n)dϕ(μ).
If we now suppose that dim M = dim N + 1 with both M and N oriented (or
simply, the fibres of ϕ are oriented) and that ϕ is submersive, then it follows
that the metric g has the form
ϕ∗h
g= + θ2
λ2
for some 1-form θ on M, non-vanishing on ker dϕ.
Let ϕ : (M 3 , g) → (N 2 , h) be a semi-conformal submersion. Then there
are three quantities that come into play when setting up a correspondence
 Explicit constructions of Ricci solitons 45

as described in the opening paragraph: the dilation λ : M 3 → R (> 0); the

mean-curvature of the fibres μ : M → T M; the integrability tensor of the hor-
izontal distribution I : T M × T M → T M given by I (X, Y ) = V[HX, HY ],
where H and V are orthogonal projection onto the horizontal and vertical
spaces, respectively. We impose further restrictions by supposing that μ is the
gradient of a function and that the above three quantities are basic: λ = λ ◦ ϕ,

μ = grad ρ with ρ = ρ ◦ ϕ, ψ := ||I ||2 = a,b ||I (ea , eb )||2 = ψ ◦ ϕ, where
{ea } is an orthonormal basis for Tx M at each x ∈ M and where we place a
“bar” over corresponding quantities on N 2 .
The two most important objects to express on M 3 are its metric g and its
Ricci curvature Ricci (g). Then, as above,
ϕ∗h
g= + θ2 ,
λ2
where θ = g(U, · ) with U a unit tangent to the fibres of ϕ. For the Ricci
curvature, we obtain [2]:
  1 1
Ricci (g) = λ2 K N +  ln λ + μ(ln λ) (g − θ 2 ) − ||I ||2 g + Lμ g
4 2
− (μ + dV ln λ)2 − (dV ln λ)2 + θ  θ, (3.12)

where K N is the Gauss curvature of N , θ = (dd∗ + d∗ d)θ is the Laplacian

on forms and  denotes the symmetrized tensor product.
We now consider the soliton flow, which, as in Section 3.2, we suppose has
a decomposition of the form:

E = X + f U,

where g(X, U ) = 0 and f is to be determined. In essence, one of the compo-

nents of the soliton equation becomes d(fρ) = . . . where the right hand-side
is independent of f , so that the existence of f becomes an application of the
Poincaré Lemma (see [2] for details). We then make the assumption that X is
a gradient with basic potential: X = grad ln ν, where ν = ν ◦ ϕ.
It is a straightforward calculation to show that  := dθ = −g(U, I ) −
μ ∧ θ and, under the assumption that μ is a gradient, the 2-form  =
−g(U,&I ) is always basic (cf. [2] Lemma 3.1). Specifically,   = ϕ ∗ , where
2
 = ( 2ψ/λ )μN and where μN is the volume form on N . Thus in order to
construct θ we are required to solve the exterior differential equation:

.
dθ + μ ∧ θ = 

We are now ready to state an improved version of the ansatz given in [2].
46 P. Baird

Theorem 3.3.1 Let (N 2 , h) be a Riemannian surface and let ρ, ν, ζ : N → R

satisfy the following system of equations:
⎧
⎪
⎪ (i)
2
Ricci N + ∇ 2 ln ν − (d ln ρ)2 + (A − ζ )h = 0
⎪
⎪
⎨ 2
(ii)  ln ρ − h(grad ln ρ, grad ln ν) + ζ + B = 0
−1 (3.13)
⎪
⎪ (iii)  ln(ρ 2
νζ ) + |grad ln ρ| 2
− |grad ln ζ | 2
⎪
⎪
⎩ 2
+ h(grad ln ζ , grad ln ν) + ζ + A = 0

where A and B are constants and where equation (iii) is vacuous if ζ ≡ 0. Set
M = N × (−δ, δ)√for some δ > 0, and let ϕ : M → N be the canonical pro-
 = ϕ ∗ , ρ = ρ ◦ ϕ. Let θ be a 1-form satisfying
jection. Let  = 2 ζ μN , 

,
dθ + d ln ρ ∧ θ =  (3.14)

which is non-vanishing on ker dϕ. Write g = ϕ ∗ h + θ 2 . Then (M 3 , g) is a Ricci

soliton.

Proof : By [2], it is sufficient to prove the equivalence of the above system with
the following set of equations:
⎧
⎪
⎪
2
(i) (a) K N + 12 N ln(λ ν) − |grad ln ρ|2 + A−ψ =0
⎪
⎪ 2
⎪
⎨ (i) (b)
λ
∇
(
2
ln ν + 2d ln λ  d ln ν − (d ln ρ) 2
= αh )(some α : N → R)
 2
⎪
⎪ (ii) λ  ln ρ − h(grad ln ρ, grad ln ν) + 2 = const.
N ψ+A
⎪
⎪ λ
⎪
⎩ (iii) −1/2
N ln(ρ 2 νψ ) + |grad ln ρ|2 − 1 |grad ln ψ|2
4
+ 12 h(grad ln ψ, grad ln ν) + ψ+A
2 = 0, (3.15)
λ

where (iii) is vacuous whenever ψ ≡ 0. However, this system is conformally

−2
invariant (see [2]), and so we can replace h by  h = λ h and so, without loss
of generality, we can suppose that λ ≡ 1. Note that by doing this we break the
2
conformal invariance. We also replace ψ ≥ 0 by ζ determined by ζ = ψ (now
there is no restriction on the sign of ζ ). We claim that (i) is equivalent to (i) (a)
and (i) (b).
2
Since Ricci N = K N h, then (i)⇒(i) (b) with α = ζ − A − K N . Now the
trace of (i) implies that

2K N + δ ln ν − |grad ln ρ|2 + 2A − 2ζ

which is equivalent to (i) (a). Thus (i) implies (i) (a) and (b). Conversly, given
(i) (a) and (b), taking the trace of (i) (b) gives

2α =  ln ν − |grad ln ρ|2 ,
 Explicit constructions of Ricci solitons 47

2
which by (i) (a) equals −2K N − 2A + 2ζ , so that
2
∇ 2 ln ν − (d ln ρ)2 + K N h + AH − ζ h = 0,
which is precisely (i). 
Example 3.3.2 (Case of integrable horizontal distribution) If ζ ≡ 0, then the
horizontal distribution of ϕ is integrable and equation (3.13)(iii) becomes vac-
uous (cf. [2]). We then have the following coupled system to solve:

(i) Ricci N + ∇ 2 ln ν − (d ln ρ)2 + Ah = 0
(ii)  ln ρ − h(grad ln ρ, grad ln ν) + B = 0 .
For example, the dynamical system of Theorem 3.2.1 can be seen to arise this
way.
Example 3.3.3 (Case of minimal fibres–the geometry Nil) The fibres of ϕ
are minimal if and only if ρ = const. In this case equation (ii) implies that
2
ζ = −B is constant and (i) becomes:
Ricci N + ∇ 2 ln ν + (A − B)h = 0 ,
which is the equation for a gradient soliton on the surface (N 2 , h). If ζ = 0,
then the horizontal distribution is integrable and we locally have a product
structure M 3 =  × R, where  is a 2-dimensional gradient soliton.
If on the other hand ζ = 0, then equation (iii) implies that
2
 ln ν + ζ + A = 0 ,
which combined with the trace of (i) gives the identity:
2
2K N + A − 3ζ = 0 ,
so that in particular the curvature K N must be constant. If we√take N = R2 with
its canonical metric and coordinates (x, y) and set ζ = 1/ 2, then, applying
the construction of Theorem 3.3.1, we obtain   = dx ∧ dy. To find θ we are
required to solve
dθ = dx ∧ dy .
so that, up to a diffeomorphism, we can take θ = xdy + dz (recalling that θ
must not vanish on vertical vectors). This then gives the soliton metric:
g = dx 2 + dy 2 + (xdy + dz)2 ,
which is the metric of the geometry Nil. It is easy to calculate the soliton flow,
which is given by
∂ ∂ ∂
E = −x −y − 2z .
∂x ∂y ∂z
This is not a gradient with respect to the metric g [2].
48 P. Baird

3.4 Higher dimensional Ricci solitons via projection

In general dimensions, the problem of understanding the soliton equations
in terms of the parameters of a semi-conformal submersion ϕ : M m → N n
become more difficult. This is because the fundamental tensors of a submersion
come into play when expressing the Ricci curvature on M m in terms of that on
N n , see [3]. However, provided the fibres have dimension 1 and ϕ is harmonic,
i.e. a harmonic morphism, then we still obtain a manageable expression.

Proposition 3.4.1 Let ϕ : M n+1 → N n be a submersive harmonic morphism,

then
n(n − 2)
Ricci M = ϕ ∗ Ricci N +  ln λg H − d ln λ2
2
+ n d(U (ln λ)) + d∗  + n(n − 2)U (ln λ)d ln λ  θ
n2 1 1
− U (ln λ)2 + ||||2 θ 2 − (ei , · )(ei , · ) ,
2 4 2

where {ei } is an orthonormal basis of the horizontal space (ker dϕ)⊥ and
g H = ϕ ∗ h/λ2 is the horizontal part of the metric.

Proof : As before, let U be a unit vector field tangent to the fibres of ϕ and let
θ = g(U, · ). If one rescales: V = λn−2 U ,   = d
θ = λ−n+2 θ ,  θ , then in [3],
the Ricci curvature is computed on components as follows; here we suppose
that X, Y are orthogonal to U :

Ricci M (U, U ) = −(n − 2) ln λ + 2(n − 2)U (U (ln λ))

||2
− n(n − 1)||Vgrad ln λ||2 + 14 λ2n−4 ||
Ricci (X, U ) = (n − 1)X(U
M
 (ln λ)) 
(X) − 2(n − 2)
+ 12 λn−2 d∗  (grad ln λ, X)
Ricci M (X, Y ) = Ricci N (dϕ(X), dϕ(Y )) + X, Y  ln λ
 , iY 
− (n − 1)(n − 2)X(ln λ)Y (ln λ) − 1 λ2n−4 iX   ,
2

where iX denotes contraction with respect to the vector X. Now

 + (n − 2)d ln λ ∧ θ ,
 = λn−2 

from which, after some calculation, we obtain

 − (n − 2)λn−2 
d∗  = λn−2 d∗  (grad ln λ, · )
(
+ (n − 2) −  ln λ − nU (ln λ)2 + (n − 2)||Hgrad ln λ||2 θ
)
+ d(U (ln λ)) .
 Explicit constructions of Ricci solitons 49

On substituting this and evaluating on different combinations of horizontal and

vertical vectors, the formula of the proposition follows. 

We are now able to obtain the soliton equations in terms of the parameters
of a semi-conformal submersion as expressed by the following theorem.

Theorem 3.4.2 Let ϕ : M n+1 → N n be a submersive harmonic morphism.

Then M n+1 admits a soliton structure with flow E = X + f U where g(X, U ) =
0 and with constant A if and only if:
n(n − 2)
0 = ϕ ∗ Ricci N +  ln λ − f U (ln λ) g H − d ln λ2
 2 
+ nd(U (ln λ)) + d∗  + n(n − 2)U (ln λ)d ln λ + df − (n − 2)f d ln λ  θ
 
n2 1
+ (n − 2)f U (ln λ) − U (ln λ)2 − ||||2 θ 2
2 4
1 1
− (ei , · )(ei , · ) + LX g + Ag ,
2 2
where {ei } is an orthonormal basis for (ker dϕ)⊥ .

In order to obtain the above formula, it suffices to note that

Lf U g = −2f U (ln λ)g H + 2(df − (n − 2)f d ln λ)  θ + 2(n − 2)f U (ln λ)θ 2 ,

and to then apply the soliton equation (3.2).

Example 3.4.3 (The Bryant solitons) We consider M n+1 = N n × J endowed

with a metric of warped propduct type: g = (h/λ2 ) + dt 2 , where J is an open
interval in R with its coordinate t and where λ = λ(t). We take ϕ to be the
canonical projection ϕ : M → N . Then the fibres are geodesic and ϕ is both
semi-conformal and harmonic. Note also that the horizontal distribution is
integrable, so that  vanishes. We will suppose that the horizontal component
X of the soliton flow E vanishes and that N is Einstein, with Ricci N = ah, so
that ϕ ∗ Ricci N = aλ2 g H . Then the equation of the above theorem is equivalent
to the pair of equations

0 = aλ2 +  ln λ − f U (ln λ) + A
0 = U (f ) + nU (U (ln λ)) − nU (ln λ)2 + A .

Now suppose that t denotes a unit speed parameter along the fibres, so that
U = ∂/∂t and θ = dt, then a necessary consequence of the equations is that
f = f (t). There are product solutions given by λ =const. If, on the other hand,
we suppose that λ is non-constant and work on a neighbourhood where λ = 0,
50 P. Baird

then, on letting {ea } denote a local orthonormal frame, we obtain

 ln λ = Tr ∇ 2 ln λ = d ln λ(∇ea ea ) + U (U (ln λ) = −n((ln λ) )2 + (ln λ) .

We thereby obtain the system:


0 = λ2 a + (ln λ) − n((ln λ) )2 − f (ln λ) + A

0 = f + n(ln λ) − n((ln λ) )2 + A
One can eliminate f from these equations to obtain a 3rd order ODE in
λ, solutions of which correspond to soliton metrics (see [2] for the three-
dimensional case). The Bryant solitons arise by taking A = 0 and making the
substitutions: ρ = 1/λ, x = ρ  , y = ρf + nρ  , ds = dt/ρ, thus leading to a
system of the form: x  (s) = F (x, y), y  (s) = G(x, y) for appropriate F and
G. In particular, on taking N to be the n-sphere with its canonical metric, one
obtains a solution corresponding to a complete rotationally symmetric steady
soliton [4].

3.5 The 4-dimensional geometry Nil4

If a metric has sufficient symmetry, it may be possible to solve the soliton equa-
tions directly. This was done in [2] in order to show non-existence of any soliton
structure on the geometry SL *2 (R) and uniqueness on the other 3-dimensional
geometries. Here we perform the calculations for the 4-dimensional geometry
Nil4 and so produce an explicit coordinate expression for the soliton flow as
well as showing its uniqueness.
The geometry Nil4 can be identified with R4 endowed with the metric
2
x1 2
g = dx1 2 + dx3 2 + (dx2 + x1 dx3 )2 + dx4 + x1 dx2 + dx3 .
2
This can be calculated from its characterisation as a left-invariant metric with
respect to the group structure of Nil4 . There is a natural harmonic Rieman-
nian submersion (x1 , x2 , x3 , x4 ) → (x1 , x2 , x3 ) onto (R3 , h = dx1 2 + dx3 2 +
(dx2 + x1 dx3 )2 ) = Nil3 .
The components of the inverse matrix (g ij ) = (gij )−1 are given by
⎛ ⎞
1 0 0 0
⎜ ⎟
⎜0 −x1 −x1 1 + x21 ⎟
2
1 + x1 2
⎜ ⎟
⎜ x1 2 ⎟
⎜0 −x1 1 ⎟
⎝ 2
2 ⎠
x1 2 x1 2 x1 2
0 −x1 1 + 2 2
1+ 2
 Explicit constructions of Ricci solitons 51

from which we can compute the non-zero Christofell symbols:

3x1 2 x1 2
1
22 = −x1 , 1
23 = − 12 1 + 2
1
, 24 = − 12 , 1
33 = −x1 1 + 2
,
2
1
34 = − x21 , 2
13 = 1
2
1− x1
2
, 2
14 = 12 , 3
12 = 12 ,
x1 2
3
13 = x1
2
, 4
12 = 1
2
1− 2
, 4
14 = − x21 ,

Then we compute the Ricci curvature, which is given by:

1 1 x1 2 2
Ricci = −dx1 2 − dx3 2 + x1 dx2 + dx3 + dx4 .
2 2 2
We now introduce the soliton flow E = α∂1 + β∂2 + γ ∂3 + f ∂4 , compute
LE g and substitute into the soliton equation (3.2). We omit the details of this
latter calculation. When we equate the different coefficients to zero, we obtain
the following set of equations to solve:

dx1 2 : −2 + 2∂1 α + 2A = 0
x1 2
dx1 dx2 : 2∂2 α + 2(1 + x1 2 )∂1 β + 2x1 1 + 2
∂1 γ + 2x1 ∂1 f = 0
2
x1 2 x1 2
dx1 dx3 : 2∂3 α + 2x1 1 + 2
∂1 β + 2 1 + 2
∂1 γ + x1 2 ∂1 f = 0
dx1 dx4 : 2∂4 α + 2x1 ∂1 β + x1 2 ∂1 γ + 2∂1 f = 0
x1 2
dx2 2 : x1 2 + 2x1 α + 2(1 + x1 2 )∂2 β + 2x1 1 + 2
∂2 γ
+ 2x1 ∂2 f + 2(1 + x1 2 )A = 0
x1 2
dx2 dx3 : x13 + (2 + 3x1 2 )α + 2(1 + x1 2 )∂3 β + 2x1 1 + 2
∂3 γ + 2x1 ∂3 f
2 2 2
+ 2x1 1+ x21 ∂2 β+2 1+ x21 ∂2 γ +x1 2 ∂2 f + 2A(2x1 +x1 3 ) = 0
x1 2
dx2 dx4 : 2x1 + 2α + 2(1 + x1 2 )∂4 β + 2x1 1 + 2
∂4 γ + 2x1 ∂4 f
+ 2x1 ∂2 β + x1 2 ∂2 γ + 2∂2 f + 4x1 A = 0
x1 4 x1 2 x1 2
dx3 2 : −1 + 4
+ 2x1 1 + 2
α + 2x1 1 + 2
∂3 β
2 2
x1 2 x1 2
+2 1 + 2
∂3 γ + x1 2 ∂3 f + 2A 1 + 2
=0
x1 2 x1 2
dx3 dx4 : x1 2 + 2x1 α + 2x1 1 + 2
∂4 β + 2 1 + 2
∂4 γ + x1 2 ∂4 f
+ 2x1 ∂3 β + x1 2 ∂3 γ + 2∂3 f + 2Ax1 2 = 0
dx4 2 : 1 + 2x1 ∂4 β + x1 2 ∂4 γ + 2∂4 f + 2A = 0 .
52 P. Baird

We now make the substitution:

⎛ ⎞ ⎛ 2 2 2
⎞⎛ ⎞
u 2x1 1 + x21 2 1 + x21 x1 2 β
⎝v⎠=⎜ ⎝ 2
⎟⎝ ⎠
⎠ γ ⇒
2x1 x1 2
w 1 x1 0 f
⎛ ⎞ ⎛ ⎞⎛ ⎞
β −2x1 x1 3 4 + 4x1 2 u
⎝γ ⎠ = 1 ⎜ ⎝ 2 −x1 2 −4x1
⎟⎝ ⎠
⎠ v
4 4
f x1 2 2 − x21 −4x1 − 2x1 3 w
This leads to the revised set of equations:
⎧
⎪
⎪ (i) ∂1 α − 1 + A = 0
⎪
⎪ 2 α + x1 ∂1 v + 2∂1 w − u + 2 x1 v = 0
1 2
⎪
⎪ (ii) 2∂
⎪
⎪
⎪
⎪
2
(iii) 2∂3 α + ∂1 u − x1 u − x1 1 − x21 v − (2 − x1 2 )w = 0
⎪
⎪
⎪
⎪
⎪
⎪ (iv) 2∂4 α + ∂1 v − 2w = 0
⎪
⎨ (v) x ∂ v + 2∂ w + 2x α + x 2 + 2(1 + x 2 )A = 0
1 2 2 1 1 1
⎪
⎪ (vi) x1 ∂3 v + 2∂3 w + ∂2 u + x1 3 + (2 + 3x1 2 )α + 2x1 (2 + x1 2 )A = 0
⎪
⎪
⎪
⎪ (vii) x1 ∂4 v + ∂2 v + 2∂4 w + 2α + 2x1 + 4Ax1 = 0
⎪
⎪
⎪
⎪ 4 2 2
⎪
⎪ (viii) ∂3 u + x1 (2 + x1 2 )α − 1 + x41 + 2 1 + x21 A=0
⎪
⎪
⎪
⎪ (ix) ∂4 u + ∂3 v + 2x1 α + x1 + 2Ax1 = 0
2 2
⎪
⎩
(x) ∂4 v + 1 + 2A = 0
Then (i), (ii), (iv) and (x) give
α = (1 − A)x1 + p
v = −(1 + 2A)x4 + q
1 1
w = ∂4 α + ∂1 v = ∂4 p + ∂1 q
2 2
1
u = 2∂2 α + x1 ∂1 v + 2∂1 w + x1 2 v
2
1 2 x1 2
= 2∂2 p + x1 ∂1 q + ∂11 q − x1 x4 (1 + 2A) + q,
2 2
where p = p(x2 , x3 , x4 ) and q = q(x1 , x2 , x3 ). The remaining equations now
impose the following conditions on p and q:
(a) ∂111 q − 2x1 ∂2 p + 2∂3 p − (2 − x1 2 )∂4 p = 0
(b) 2∂24 p + ∂12 q + x1 ∂2 q + 2x1 p + 3x1 2 + 2A = 0
2
(c) ∂112 q + x1 ∂12 q + ∂13 q + 2∂22 p + 2∂34 p + x21 ∂2 q
+x1 ∂3 q + (2 + 3x1 2 )p + 2x1 + 4x1 3 + (2x1 − x1 3 )A = 0
(d) 2∂44 p + ∂2 q + 2p + 3x1 = 0
2
(e) ∂113 q + 2∂23 p + x1 ∂13 q + x21 ∂3 q + x1 (2 + x1 2 )p − 1
5x1 2 x1 4
+2x1 2 + 4
+ 2− 2
A=0
(f) 2∂24 p + ∂3 q + 2x1 p + 2 x1 − x1 2 A
5 2
= 0.
 Explicit constructions of Ricci solitons 53

We obtain the following consequences:

(f) : ∂3 q + 2x1 p + 52 x1 2 − x1 2 A = −2∂24 p
∂1 (d) : ∂12 q + 3 = 0 (⇒ ∂112 q = 0)
∂1 (b) : ∂112 q + x1 ∂12 q + ∂2 q + 2p + 6x1 = 0 ⇒ ∂2 q + 2p + 3x1 = 0
(d) : ∂44 p = 0 ⇒ p = x4 s + t ,
where s = s(x2 , x3 ) and t = t(x2 , x3 ). Then
3 − 2A
(b) : 2∂2 s = 3 − 2A ⇒ s = x3 + ξ ,
2
where ξ = ξ (x3 ). Also
3 − 2A 5
∂24 p = ∂2 s = ⇒ ∂3 q = 2A − 3 − 2x1 p − x1 2 + x1 2 A
2 2
⇒ ∂13 q = −2p − 5x1 + 2x1 A → ∂113 q = −5 + 2A .

Then (e) is equivalent to

9x1 2
2∂23 t − 6 − + 4A + 3x1 2 A = 0 .
2
Taking ∂1 of this then yields A = 3/2 so that s = ξ (x3 ) and ∂23 t = 0. We
therefore have

∂2 t = σ (x2 ), p = x4 ξ (x3 ) + t(x2 , x3 ) .

for a function σ = σ (x2 ). Equation (c) now shows that σ  = −ξ  = a constant,

so that
ax2 2
σ = ax2 + b, ξ = −ax3 + c, ∂2 t = ax2 + b ⇒ t = + bx2 + e(x3 ) ,
2
for constants a, b, c and for some function e = e(x3 ). We now have
ax2 2
p = x4 (−ax3 + c) + + bx2 + e(x3 )
2
5
∂3 q = 2A − 3 − 2x1 p − x1 2 + x1 2 A
2
∂2 q = −2p − 3x1 .

Then ∂23 q = ∂32 q ⇒ x1 ∂2 p = ∂3 p ⇔ x1 (ax2 + b) = −ax4 + e (x3 ) ⇔ a =

b = 0 and e = const. Thus

p = cx4 + e and ∂2 q = −2p − 3x1 = −2cx4 − 2e − 3x1 .

But since q is independent of x4 , we must have c = 0 so that p = e. Also

∂2 q = −3x1 − 2e ⇒ q = −3x1 x2 − 2ex2 + ρ(x1 , x3 ) ,

54 P. Baird

for some function ρ = ρ(x1 , x3 ), so that

∂3 q = ∂3 ρ = −2ex1 − x1 2 ⇒ ρ = −2ex1 x3 − x1 2 x3 + r(x1 ) ,

for some function r = r(x1 ). But now

ax1 2
(a) : ∂111 q = 0 ⇒ r  = 0 ⇒ r = + bx1 + c ,
2
for some (new) constants a, b, c. We conclude that

p=e
2
q = −3x1 x2 − x1 2 x3 − 2ex2 − 2ex1 x3 + ax21 + bx1 + c
α = − x21 + e
ax 2
v = −4x4 − 3x1 x2 − x1 2 x3 − 2ex2 − 2ex1 x3 + 2[ + bx1 + c
w = 12 (−3x2 − 2x1 x3 − 2ex3 + ax1 + b)
u = x1 (−3x2 − 2x1 x3 − 2ex3 + ax1 + b) − 2x3 + a − 2x1 2 x4
2 ax 2
+ x21 (−3x1 x2 − x1 2 x3 − 2ex2 − 2ex1 x3 + 2[ + bx1 + c) ,

for constants a, b, c, e, which gives

3x2 x1
β=− + a − 2ex3 + b
2 2
a
γ = −x3 +
2
f = −2x4 − ex2 .

It is easily seen that the arbitrary constants correspond to the addition of a

Killing vector field to E, so we can summarize the above calculations in the
following theorem.

Theorem 3.5.1 The geometry Nil4 admits a unique soliton structure, with flow
defined up to addition of a Killing vector field, given by

x1 3x2
E=− ∂1 − ∂2 − x3 ∂3 − 2x4 ∂4 .
2 2
This result is confirmed by Theorem 3.4.2, where we take ϕ : Nil4 → Nil3
to be the natural projection and exploit the soliton structure on Nil3 given
by Example 3.3.3; indeed, we apply Theorem 3.4.2 with λ ≡ 1, θ = dx4 +
2
x1 dx2 + x21 dx3 and X = − x21 ∂1 − 3x22 ∂2 − x3 ∂3 . However, 3.4.2 does not
establish uniqueness.
 Explicit constructions of Ricci solitons 55

References
[1] P. Baird, A class of three-dimensional Ricci solitons, Geometry and Topology 13
(2009), 979–1015.
[2] P. Baird and L. Danielo, Three-dimensional Ricci solitons which project to surfaces,
J. reine angew. Math., 608 (2007), 65–91.
[3] P. Baird and J. C. Wood, Harmonic Morphisms between Riemannian Manifolds,
London Math. Soc. Monograph (New Series), vol. 29, Oxford University Press,
2003.
[4] R. L. Bryant, Ricci flow solitons in dimension three with SO(3)-symmetries,
preprint, Duke Univ., Jan 2005.
[5] B. Chow, S-C. Chu, D. Glickenstein, C. Guenther, J. Isenberg, T. Ivey, D. Knopf,
P. Lu, F. Luo and L. Ni, The Ricci flow: Techniques and Applications, Part 1:
Geometric aspects, AMS Mathematical Surveys and monographs, 135, 2007.
[6] B. Chow and D. Knopf, The Ricci flow: An Introduction, Mathematical Surveys
and Monographs, Vol. 110, American Mathematical Society, Providence, RI, 2004.
[7] C. Guenther, J. Isenberg and D. Knopf, Stability of the Ricci flow at Ricci-flat
metrics, Comm. Anal. Geom. 10 (2002), no. 4, 741–777.
[8] C. Guenther, J. Isenberg and D. Knopf, Stability of Ricci nilsolitons, preprint
(2006).
[9] R. Hamilton, A compactness property for solutions of the Ricci flow, Amer. J.
Math. 117 (1995), 545–572.
[10] T. Ivey, Ricci solitons on compact three-manifolds, Differential Geom Appl. 3 (4)
(1993), 301–307.
[11] T. Ivey, New examples of complete Ricci solitons, Proc. Amer. Math. Soc. 122
(1994), 241–245.
[12] J. Lauret, Ricci soliton homogeneous nilmanifolds, Math. Ann. 319 (2001), 715–
733.
[13] J. Lott, On the long-time behaviour of type-III Ricci flow solutions, Math. Annalen,
339, No. 3 (2007), 627–666.
[14] G. Perelman, The entropy formula for the Ricci flow and its geometric applications,
arXiv:math.DG/0211159.
[15] N. Sesum, Linear and dynamical stability of Ricci flat metrics, Duke. Math. J.,
133 (2006), 1–26.

Author’s address:

Département de Mathématiques,
Université de Bretagne Occidentale,
6 Avenue Le Gorgeu, – CS 93837
29238 Brest,
France
[email protected]
 4
Open Iwasawa cells and applications
to surface theory
josef f. dorfmeister

4.1 Introduction
In recent years, many surfaces of a special type, like surfaces of constant mean
curvature (CMC) in R3 , Willmore surfaces in S n or spacelike mean curvature
surfaces in Minkowski space L3 , have been constructed using loop groups.
The procedure in all these cases is fairly similar: one considers a ‘Gauss type
map’ from a Riemann surface M to some real symmetric space G/K, like
the classical Gauss map or a conformal Gauss map, and characterizes a class
of surfaces by the harmonicity (or the conformal harmonicity) of this Gauss
map. Lifting the Gauss map to a map F from the universal cover M̃ into G
we obtain a ‘moving frame’. It has been shown in [7] how all such frames
can be constructed from holomorphic data: considering the moving frame for
each member of the associated family one obtains an ‘extended frame’ Fλ and
one can show that it suffices to construct all such extended frames, since, as an
added feature, for all these special surface classes there exists a simple formula,
‘Sym type formula’, which reconstructs the given immersion from its extended
frame.
The heart of the loop group method for the construction of surfaces thus
is a procedure that produces all extended frames of all surfaces of the special
classes to which this method applies. This procedure involves loop group
decompositions. For simplicity, let’s consider conformal immersions like CMC
and spacelike CMC from some simply connected domain D ⊂ C into R3 and
L3 respectively. Then the extended frame F has, for all z ∈ D \ S, S a discrete
subset of D, a ‘Birkhoff decomposition’ F = F− L+ . (Here F− only contains
non-positive powers of λ and starts with I and L+ only contains non-negative
powers of λ in a Fourier expansion.) The matrix function F− maps into GC
and is meromorphic with poles at the points of S. The Maurer-Cartan form
η = F−−1 dF− of F− is a meromorphic (1, 0)− form and is called the ‘normalized

56
 Open Iwasawa cells 57

potential’ for the given immersion. It is of the form η = λ−1 η−1 (z)dz and has
values in pC , where g = k + p is the Cartan decomposition of the symmetric
space G/K. As a consequence, if one wants to construct an extended frame
for a surface in one of the classes under consideration, one will start with
some normalized potential η = λ−1 η−1 (z)dz, a meromorphic (1, 0)−form with
values in pC , which has a meromorphic solution to the ODE dF− = F− η. In
the next step one decomposes F− in the form F− = F W+ , with F ∈ G. If G is
maximal compact in the complex Lie group GC , then this decomposition always
exists and is called ‘Iwasawa decomposition’. But in many cases relevant to
surface theory G is non-compact. In the finite dimensional case such a situation
has been investigated by Aomoto, Matsuki and Rossman [1], [13], [15]. In the
(infinite dimensional) loop group case similar results have been obtained by
Kellersch [10], [11]. It turns out that a decomposition of the form F− = F W+
only exists away from a singular set S. This singular set may consist of points,
but may also contain curves. The matrix function F is, where defined, the
extended frame of a harmonic map associated with surfaces of our class. Then
a ‘Sym formula’ produces the actual surface. More precisely, the Sym formula
produces a map, called here ‘weak immersion’, which is an immersion of the
desired surface type, wherever its differential has rank two.
In the construction outlined above, three types of singularities occur: firstly,
poles in the normalized potential. Secondly, singularities due to the non-global
Iwasawa splitting and, thirdly, branch points or branching curves, where the
differential of the mapping defined by the Sym formula drops rank.
The first kind of singularities can be taken care of by considering a slightly
different construction, producing holomorphic potentials (with a less trivial
Fourier expansion). The last kind of singularities is sometimes unavoidable
and will not be discussed in this note.
Singularities of the second kind, stemming from the non-globality of the
Iwasawa splitting for non-compact G, have been investigated by a few authors
only so far and for G = Sl(2, R) only, to the best of our knowledge. It turns
out [2] that the coefficients of the frame F have strong singularities along the
singular set S, while the actual weak immersion may be smooth across S with
a differential of rank lower than two.
To understand this behaviour better, we consider general symmetric spaces
+
G/K and the corresponding Iwasawa decompositions LGC σ = δ∈ LGσ · δ ·
L+ GC σ , where  is a set of representatives for the obvious action of LGσ ×
+ C C
L Gσ on LGσ . These cosets will be called ‘Iwasawa cells’. Similar to what
was found in the Sl(2, R) case one also finds, for general real G, that in general
several open Iwasawa cells exist. Clearly, considering F− as above, we obtain
a well-behaved extended frame of some harmonic map as long as F− stays in
58 J. F. Dorfmeister

an open Iwasawa cell. We are therefore interested in a description of the open

Iwasawa cells and their boundary.
As main result of this paper we prove a first step in this direction:

Theorem 4.1.1 (1) The union of all open Iwasawa cells is dense in LGC σ.
(2) There exists a line bundle L∗ over LGC
σ and a real analytic section α∗
∗ C ∗
of L such that some g ∈ LGσ is in an open Iwasawa cell only if α (g) = 0.

In applications to surface theory we will find real analytic, complex valued

functions (not only sections of some bundle) such that these functions do not
vanish if F− (z, λ) is in an open Iwasawa cell.

4.2 Basic notation and the Birkhoff decomposition

Let G be a real form of the simply connected, complex, semi-simple, matrix Lie
group GC . Then we consider the group LGC of loops in GC , i.e. the set of all
maps from the unit circle S 1 into GC , where we assume that each matrix entry
has a Fourier expansion for which the sum of absolute values of its coefficients,
multiplied with a weight, converges (weighted Wiener topology). For more
details we refer to [4].
Let τ denote the anti-holomorphic involution of GC with G = F ix(τ ).
In this paper we consider exclusively inner symmetric spaces G/K. We can
assume that there exists a holomorphic involution σ such that K = F ix(σ )
and that σ and τ commute. For a description of harmonic maps into symmetric
spaces it is advantageous to use ‘twisted loop groups’.
The following groups will be used throughout this paper.

1 LGC C
σ = {g ∈ LG ; σ (g(λ)) = g(−λ)},
2 L Gσ = {g ∈ LGC
+ C
σ ; g has a holomorphic extension to the open unit disk},
3 L+ G
∗ σ
C
= {g ∈ LG C
σ ; g has a holomorphic extension to the open unit disk
and g(0) = I },
4 L− GC C
σ = {g ∈ LGσ ; g has a holomorphic extension to the upper
hemisphere},
5 L− C C
∗ Gσ = {g ∈ LGσ ; g has a holomorphic extension to the upper hemisphere
and g(∞) = I },
6 LGσ = {g ∈ LGC ; g(λ) ∈ G for all λ ∈ S 1 }.

Note, we identify the unit disk with the lower hemisphere of the Riemann
sphere S 2 . And holomorphicity in the upper hemisphere means that after chang-
ing λ → λ−1 one has a holomorphic quantity on the open unit disk.
 Open Iwasawa cells 59

Using this notation we can state the first important decomposition theorem
(see e.g. [14],[3]).

Theorem 4.2.1 (Birkhoff decomposition Theorem)

,
LGC σ = L− GC + C
σ · ω · L Gσ (4.1)
ω∈

In particular, every g ∈ LGC

σ can be written in the form

g = g− ωg+ (4.2)

for some g∗ ∈ L∗ GC σ and ω ∈ . The set  for the disjoint union above is
closely related to the Weyl group of LGCσ (see e.g. [14]).
Moreover, the group multiplication L− GC + C C
σ × L Gσ → LGσ is an analytic
C
diffeomorphism onto an open dense subset of LGσ .

Remark 4.2.2 (1) The Birkhoff decomposition for g in the ‘big (left Birkhoff)
cell’
L− GC + C − C
σ · L Gσ can be made unique if one requires g− ∈ L∗ Gσ or g+ ∈
+ C
L∗ Gσ .
(2) For the other cells one can also formulate a condition that makes the
representation unique [9],[14].

4.3 Iwasawa decomposition

Let GC , τ and G be as before. Then we consider the group action

(LGσ × L+ GC C C
σ ) × LGσ → LGσ , where (g, v+ ).h = ghv+
−1
(4.3)

Let  denote a set of representatives for the orbits (= double cosets) of

this group action. Then we obtain a general form (i.e. without any precise
description of ) of the “Iwasawa decomposition”.

Theorem 4.3.1 (Iwasawa Decomposition)

,
LGCσ = LGσ · δ · L+ GC
σ (4.4)
δ∈

In particular, every g ∈ LGC

σ can be written in the form

g = hδg+ (4.5)

for some h ∈ LGσ , g+ ∈ L+ GC

σ and δ ∈ .
60 J. F. Dorfmeister

Remark 4.3.2 (1) The finite dimensional case of an Iwasawa decomposition

for any real form G of some complex semisimple Lie group GC has been
investigated by Aomoto [1]. More general situations have been considered by
Matsuki [13] and Rossman [15]
(2) The case of untwisted loop groups has been investigated by Kellersch
[10], [11].

4.4 Iwasawa decomposition via Birkhoff decomposition

Assume that G is maximal compact in GC , then the Iwasawa Decomposition
in LGC is global, i.e. for every g ∈ LGC there exist h ∈ LG and v+ ∈ L+ GC
such that g = hv+ . This is equivalent to that for every g ∈ LGC there exists
some v+ ∈ L+ GC σ such that τ (g)−1 g = τ (v+ )−1 v+ holds. In particular, for
every g ∈ LGC the matrix τ (g)−1 g is in the big Birkhoff cell.
What we just explained works equally well in the case of twisted loop
groups.
The trick to consider a Birkhoff decomposition of τ (g)−1 g also works (to
some extent–see below) in the case of arbitrary real forms G of GC . However,
in the general case it is not possible to prove that τ (g)−1 g is in the big Birkhoff
cell for every g ∈ LGC −1
σ . And even if τ (g) g is in the big Birkhoff cell, it does
not follow that τ (g) g = τ (v+ ) v+ holds for some v+ ∈ L+ GC
−1 −1
σ.
Below we will discuss this situation for twisted loop groups. The case of
an untwisted loop group is very similar. Since it has been investigated in much
detail by Kellersch [10],[11] we will not address it in this note.
Starting from the expression τ (g)−1 g one tries to simplify it by trans-
formations of the type τ (g)−1 g → τ (v+ )(τ (g)−1 g)v+ −1 as much as possi-
ble (we will explain below how we interpret this statement). Once one has
obtained this way an expression, say β, that cannot be simplified any more, one
writes β = τ (δ)−1 δ (what can be done by construction) and altogether obtains
gv+ −1 δ −1 = τ (gv+ −1 δ −1 ). Calling this expression h one obtains the Iwasawa
decomposition g = hδv+ of g.
In view of the equality τ (v+ ) · (τ (g)−1 g) · v+ −1 = β we consider decompo-
sitions of the form τ (g)−1 g = τ (v+ )−1 · β · v+ . This reminds one of a Birkhoff
decomposition. However, since we require here that the element in L− GC σ is
related to the element in L+ GC σ , it is, a priori, not clear, that β can be chosen
as a representative of a Weyl group element.

Theorem 4.4.1 ([10],[11], Theorem 3.64) Let GC be a simply-connected, semi-

simple, complex matrix Lie group and τ an anti-holomorphic involution with
fixed point group G. Assume that G/K is an inner symmetric space, where
 Open Iwasawa cells 61

K = F ix(σ ) for some inner involution σ of G. Let U be a maximal compact

subgroup of GC which is invariant under σ and τ and T a maximal torus in U
which is fixed pointwise by σ . Let N denote the group of Laurent polynomials
in LGCσ which normalize T .
Then for every g ∈ LGC + C
σ there exists some v+ ∈ L Gσ and some n ∈ N
such that

τ (g)−1 g = τ (v+ )−1 · n · v+ . (4.6)

In particular, we have n ∈ LGC

σ.

In [10],[11] and in the untwisted case the corresponding result is Proposition

3.36 and the choice of n is specified much further. In the twisted case no similar
detailed description of the ‘middle term’  has been given yet in a general
setting. However, we will make a first step towards a detailed description of the
open Iwasawa cells.
For this we note that Lemma 4.37, Proposition 4.38 and Proposition 4.39 of
[10],[11] still hold in the twisted case.

Theorem 4.4.2 With the notation of the last theorem we can assume w.l.g.
that n has the form n = qt, where q is in the normalizer of T in U and t is a
homomorphism from S 1 into T . In particular, n is a representative of a Weyl
group element of the twisted loop group LGC σ.

Proof Consider the automorphism σ̂ of LGC given by σ̂ (g)(λ) = σ (g(−λ)).

Clearly, LGC σ = F ix(σ̂ ). Kellersch’s result above, Theorem 4.1, simply states
that the stated decompositions are valid with all occurring matrices fixed by σ̂ .
It thus suffices to verify that the three results quoted above and proven in
[10],[11] for the untwisted case still hold with all occurring matrices fixed under
σ̂ . For the generalization of loc.cit Lemma 4.37 one observes that n = σ̂ (n) =
σ̂ (q) · σ̂ (exp(H− )) · σ̂ (t) and that σ̂ maps q, exp(H− ) and t onto matrices with
the same properties as stated in loc.cit. Therefore, by the uniqueness result for
the representation (in the untwisted case), we obtain that q, exp(H− ) and t
are fixed under σ̂ . The equations of loc.cit. Proposition 4.38 are valid anyway
for q, exp(H− ) and t. And the same statement holds for the proof of loc.cit
Proposition 4.39. 

Corollary 4.4.3 An element g ∈ LGC σ is in the open Iwasawa cell LGσ ·

L Gσ if and only if t(λ) ≡ I and q = τ (b)−1 b for some b ∈ K C .
+ C

Proof Note that g = hv+ is equivalent with τ (g)−1 g = τ (v+ )−1 v+ . The real-
ization τ (g)−1 g = τ (v+ )−1 · qt · v+ shows that g is as required only if τ (g)−1 g
is in the big Birkhoff cell. This happens only if qt(λ) ∈ K C . Since this matrix
62 J. F. Dorfmeister

is in particular independent of λ we obtain t(λ) ≡ I and q ∈ K C ∩ U . Now the

claim follows. 

4.5 A function defining the open Iwasawa cells

The main goal of this section is to prove that there exists a real analytic section
of some line bundle L∗ over LGC σ which contains the union of all open Iwasawa
cells in its non-vanishing set. In geometric applications we usually use loop
matrices which are real analytic on S 1 . But in our general setting we have only
considered loop matrices in some Banach algebra defined by some weighted
Wiener norm. In particular, until now the Wiener norm, given by the sum of all
absolute values of all Fourier coefficients, worked for us.
For the result below, however, we need a weighted Wiener norm, where the
weight is of the form w(r) = (1 + |nr)m with 0 < m, n.
Theorem 4.5.1 There exists a real analytic line bundle L∗ over LGC σ and a
real analytic section α ∗ of this bundle such that some g ∈ LGC
σ is in an open
∗
Iwasawa cell only if α (g) = 0.
Proof In this proof we use substantially ideas and results from [16]. First we
consider the Grassmannian Gr (n) and the group Aut(Gr (n) )) of holomorphic
automorphisms of Gr (n) , where n is given by the realization GC ⊂ Gl(n, C).
Let p : LGC → Aut(Gr (n) )) denote the map associating with a loop matrix

A(λ) = Z Aj λj the infinite matrix which has in the j −th parallel to the
diagonal the string of matrices which are all equal to Aj . Note that this is a
linear mapping. Moreover, using a weighted Wiener norm on the loop groups
and an appropriate norm on Gr (n) and on Aut(Gr (n) )), this linear map induces a
holomorphic map ([4], section 3) from LGC (n)
σ to Aut(Gr )). As a consequence,
the map A(λ) → p(τ (A)−1 A)) is real analytic. Applying the image of this
map to the base point H+ of Gr (n) we obtain a real analytic map f (A) =
p(τ (A)−1 A)).H+ from LGC (n)
σ to Gr , since the automorphism group of Gr
(n)

acts holomorphically. Next we use the holomorphic section α of the Det ∗

bundle over Gr (n) introduced in [16]. Pulling back the Det ∗ bundle along f
we obtain the real analytic bundle L∗ = f ∗ Det ∗ over LGC σ and a real analytic
section α ∗ of this pull back bundle. We know (see e.g. [7]) that α ∗ has a non-zero
value exactly at all A(λ) for which τ (A)−1 A is in the big Birkhoff cell.
Now assume that g is in an open Iwasawa cell LGσ · δ · L+ GC σ . Then
not only for the given g, but for all b in this open cell, we obtain τ (b)−1 b =
τ (u+ )−1 · ω · u+ for some u+ ∈ L+ GC −1
σ and ω = τ (δ) δ. Assume one such
element is not in the big Birkhoff cell. Then we can assume ω = qt with
 Open Iwasawa cells 63

non-trivial homomorphism t from S 1 to T . Therefore all these elements are

not in the big Birkhoff cell and therefore α ∗ vanishes for each such b. But then
the real analytic section α ∗ vanishes on an open subset of LGC σ . Since σ is
C
inner, the twisted loop group LGσ is isomorphic with the untwisted loop group
LGC . And since we have assumed that GC is simply-connected, the untwisted
loop group is connected. As a consequence, α ∗ vanishes on the full group. This
is a contradiction, since α ∗ does not vanish at I ∈ LGC σ . Therefore t ≡ I and
C
ω = q ∈ K ∩ U follows. 

Proposition 4.5.2 With the notation of the preceding Theorem, assume

α ∗ (g) = 0. Then τ (δ)−1 δ = q ∈ K C ∩ U . Moreover, if in addition LgσC =
Lgσ + δ −1 L+ gσC δ, then δ defines an open Iwasawa cell.

Proof Consider the infinitesimal stabilizer of the action of LGσ × L+ GC σ on

−1
δ. We obtain the equation hδ − δw+ = 0. This implies h = δ w+ δ, and, since
both sides are fixed by τ , a simple calculation shows that this equation is
equivalent with τ (w+ ) = q −1 w+ q. As a consequence, w+ does not contain
any λ and thus w+ = w0 ∈ K C . But then w0 = x + iy with x, y ∈ Lie(K)
and τ (w0 ) = q −1 w0 q is equivalent with q −1 xq = x and q −1 yq = −y. Since
Ad(q) only has the eigenvalues ±1, it follows that the stabilizer of the group
action on δ has the dimension of K. This is clearly also the dimension of the
action of the same group on I . For the latter case it is easy to see that the orbit
is open in LGC σ.
In view of our assumption we can conclude from the equality of the dimen-
sions of the stabilizers that also the orbit through δ is open in LGC σ. 

The proofs just given show moreover:

Corollary 4.5.3 (1) An Iwasawa cell LGσ · δ · L+ GC C

σ is open in LGσ only if
−1
the element τ (δ) δ is in the big Birkhoff cell. Under the additional assumption
in the last proposition, the converse also holds.
(2) If q is as above, then {τ (w)−1 · q · w; w ∈ K C } always has dimension
dimK.

Even if the union of all open Iwasawa cells is not equal to the non-vanishing
set of the real analytic section discussed above, we still obtain

Proposition 4.5.4 The union of all open Iwasawa cells is dense in LGC
σ.

Proof If the union of all open Iwasawa cells is not dense, then there exists an
open set in GC ∗
− σ of the complement of this union such that the section α does
not vanish anywhere on this open set. To each such point there corresponds
some q as in Proposition 4.2.
64 J. F. Dorfmeister

We note that every such q can be written in the form q = aρ, where a ∈ T
and ρ is a once and for all fixed representative of the Weyl group of K C relative
to T . Thus freedom can only come from the fact that some q = aρ, but also
some q̂ = âρ occurs. If the claim of the proposition is wrong, then there exists
some q to which uncountably many q̂  s belong. But then one of the a’s would
be a cluster point. Now (2) in the corollary above shows that this is not possible
for q  s representing different Iwasawa cells. 

Remark 4.5.5 (1) In the finite dimensional case, Aomoto [1] has given a very
satisfactory description of all occurring cells. A fairly complete description of
all Iwasawa cells has been given in the untwisted case by Kellersch [10],[11].
For the twisted case much remains to be done.

4.6 Applications to surface theory

For the general procedure we refer to the introduction. We will list a few
concrete examples.
Example 1: We consider space like surfaces in Minkowski space L3 with
inner product x, y = −x0 y0 + x1 y1 + x2 y2 . For each such immersion one
also considers a ‘Gauss map’, the future pointing unit normal vector. One
can show that the spacelike immersion has constant mean curvature (CMC) if
and only if the Gauss map is harmonic. If the immersion is defined on some
simply connected domain D, we thus need to construct all harmonic maps
from D to H2 = {v ∈ L; v, v = −1, v0 > 0}. Since H2 ∼ = SU (1, 1)/K, where
K∼ = U (1), is realized by the diagonal matrices in SU (1, 1), is a symmetric
space, the loop group method is applicable.
The complex Lie group thus is GC = Sl(2, C) with real form G = SU (1, 1)
and stabilizer group K for the symmetric space under consideration. The com-
plex anti-linear involution τ is given by τ (g) = σ3 (ḡ t )−1 σ3 , where σ3 denotes
the diagonal matrix with entries 1 and −1. The involution σ , defining the
symmetric space, is given by conjugation by σ3 .
As outlined already in the introduction, each spacelike CMC immersion can
be constructed from some “potential” η. In our case, η can be chosen of the form
η = λ−1 η−1 , where η−1 is off-diagonal and meromorphic and we can assume
that the solution to the ODE dC = Cη is also meromorphic. For convenience
we choose a base point z∗ and assume C(z∗ , λ) = I . Moreover, to make the
presentation somewhat smoother we assume that C is holomorphic in D.
Until now the setting does not only apply to spacelike CMC surfaces, but also
to CMC surfaces in R3 and timelike constant negative Gauss curvature surfaces
 Open Iwasawa cells 65

in L3 . The distinction happens, when one applies the Iwasawa decomposition

to C.
In the case of CMC surfaces in R3 the Iwasawa decomposition is global and
thus does not introduce any new singularity. But in the case under consideration,
the Iwasawa splitting is not global. As a matter of fact, fitting to the result in the
last section we have two open Iwasawa cells, defined by q = I and q = −I .
(The corresponding δ is δ = I in the first case and in the second case the
off-diagonal matrix with (1, 2) – entry λ and (2, 1) – entry −λ−1 , see e.g. [2].)
Clearly, we expect problems, where C crosses from one open cell to the
other one. We obtain in this example that the boundary between these two
cells is given by the vanishing set of the section α ∗ . One can describe this here
in more detail: We consider the map D → LSl(2, C)σ , z → C(z, λ) and pull
back the bundle L∗ . Since D is contractible, this pull back bundle is trivial and
the section α ∗ corresponds to a real analytic function F : D → C. Hence the
points in D, where the matrix function C(z, λ) touches the boundary between
the open Iwasawa cells, are exactly the points where the function F vanishes.
Let’s start now with C at z∗ . Then C = C(z, λ) starts at I . As long as z
varies such that C(z, λ) stays inside the same open Iwasawa cell as I we can
decompose C = F V+ with F ∈ LSU (1, 1)σ and V+ ∈ L+ Sl(2, C)σ .
In [2], the behaviour of F has been investigated, when z approaches the
set F ≡ 0. It was shown there that the coefficients of F behave singularly.
However, at least in some cases, the actual immersion, obtained from F by a
Sym type formula does not necessarily behave very singularly: in some cases
it stays real analytic when crossing certain parts of the set F ≡ 0, but its
differential drops rank, so that it is no longer an immersion outside of the open
Iwasawa cells.

Remark 4.6.1 In the case of quantum cohomology of CP 1 , naturally a har-

monic map of a spacelike CMC surface in L3 occurs. The main interest in this
context is to choose this surface such that its frame never leaves the open cell
it starts in.
More generally, such a requirement is imposed in tt ∗ −geometry [5].

Example 2: In this example we consider Willmore surfaces in S n . To keep

the presentation short we will be quite sketchy here. Let’s start with some con-
formal immersion y : D → S n . Then one constructs canonically some “confor-
mal Gauss map” f : D → Gr1,3 (Ln+2 ), the Grassmannian of four-dimensional
subspaces of Ln+2 with induced Minkowski metric (thus of signature (1, 3).
It is well known that the conformal Gauss map is conformally harmonic if
and only if the original conformal immersion y is a Willmore surface. Thus we
66 J. F. Dorfmeister

need to consider conformally harmonic maps from D into the symmetric space
Gr1,3 (Ln+2 ) ∼
= SO(1, n + 1)/SO(1, 3) × SO(n − 2). As a consequence, the
loop group method is applicable.
In this case the complex Lie group GC is SO(n + 2, C) with real form
∼
G = SO(1, n + 1), given by the involution τ (X) = T0 X̄T0 , where T0 is the
diagonal matrix with diagonal entries −1, 1, 1, . . . and the symmetric space
structure is given by the involution σ (B) = SBS, where S is a diagonal matrix,
with the first four entries equal to −1 and all other entries equal to 1.
Thus K C = SO(4, C) × SO(n − 2, C). The general theory we have pre-
sented in this paper does allow for many open Iwasawa cells and we believe
that they actually exist and will give rise to many effects for general Willmore
surfaces in S n . (We have not yet computed all Iwasawa cells in this example.)
However, fortunately, for the case of Willmore spheres things are very
simple. It turns out [8] that in this case the corresponding conformal harmonic
maps are of ‘finite uniton number’ and therefore the corresponding normalized
potentials can be represented as upper triangular matrices. As a consequence,
the Iwasawa splittings can be carried out explicitly and it seems that in this
case of Willmore spheres the matrix function C(z, λ) always stays in the open
Iwasawa cells which contains I . Hence the problems discussed in Example 1
will not occur for Willmore spheres (but possibly for Willmore tori).

Remark 4.6.2 Investigating CMC surfaces in H3 with mean curvature H

satisfying −1 < H < 1, see [6], one encounters a situation which is similar to
the one encountered in this note, but definitly different: the involution τ used
in [6] is not induced from an involution of some finite dimensional Lie group
(hence it is an intrinsically loop group involution). In particular, neither the work
of Kellersch nor any result of this note can be applied directly. Nevertheless one
can compute all Iwasawa cells explicitly and obtains exactly two open Iwasawa
cells. The behaviour of immersions near the boundary is analogous to Example
1. (also see [12]).

References
[1] K. Aomoto, On some double coset decompositions of complex semisimple Lie
groups, J.Math.Soc.Japan 18 (1966), 1–44
[2] D. Brander, W. Rossman, N. Schmitt, Holomorphic representation of constant
mean curvature surfaces in Minkowski space:consequences of non-compactness
in loop group methods, Adv.Math. 223 (2010), 949–986
[3] J. Dorfmeister, H. Gradl, J. Szmigielski, Systems of PDEs obtained from factor-
ization in loop groups, Acta Appl.Math. 53 (1998), 1–58
 Open Iwasawa cells 67

[4] J. Dorfmeister, Weighted l1 −Grassmannians and Banach manifolds of solutions

of the KP-equation and the KdV-equation, Math.Nachr. 180 (1996), 43–73
[5] J. Dorfmeister, M. Guest, W. Rossman, The tt ∗ structure of the quantum cohomol-
ogy of CP 1 from the viewpoint of differential geometry, Asian J.Math. 14 (2010),
417–438
[6] J. Dorfmeister, J. Inoguchi, S. Kobayashi, Constant mean curvature surfaces in
hyperbolic 3-space via loop groups, to appear
[7] J. Dorfmeister, F. Pedit and H. Wu, Weierstrass-type representations of harmonic
maps into symmetric spaces, Comm.Anal.Geom. 6 (1998), 633–668.
[8] J. Dorfmeister, P. Wang Willmore Spheres in S n , work in progress
[9] V. Kac, D. Peterson, Infinite flag varieties and conjugacy theorems,
Proc.Nat.Acad.Sci.USA 80 (1983), 1772–1782
[10] P. Kellersch, Eine Verallgemeinerung der Iwasawa Zerlegung in Loop Gruppen,
Dissertation, TU München, 1999
[11] P. Kellersch, The Iwasawa decomposition for the untwisted group of loops in
semisimple Lie groups Balkan Press 2003, http://www.mathem.pub.ro/dgds/mono/
dgdsmono.htm
[12] S. Kobayashi, Totally symmetric surfaces of constant mean curvature in hyperbolic
3-space, Bull.Aust.Math.Soc 82 (2010), 240–253
[13] T. Matsuki, The orbits of affine symmetric spaces under the action of minimal
parabolic subgroups, J.Math.Soc.Japan 31 (1979) 331–357
[14] A. Pressley, G. Segal, Loop groups, Oxford Mathematical Monographs, Oxford
University Press 1986
[15] W. Rossmann, The structure of semisimple symmetric spaces, Canad.J.Math 31
(1979), 157–180
[16] G. Segal, G. Wilson, Loop groups and equations of KdV type, Inst.Hautes Etudes
Sci.Publ.Math. 61 (1985), 5–65

Authors’ addresses:

Josef F. Dorfmeister
Fakultät für Mathematik,
Technische Universität München,
Boltzmannstr. 3,
D-85747 Garching,
Germany
[email protected]
 5
Multiplier ideal sheaves and
geometric problems
akito futaki and yuji sano

Abstract
In this expository article we first give an overview on multiplier ideal sheaves
and geometric problems in Kählerian and Sasakian geometries. Then we review
our recent results on the relationship between the support of the subschemes cut
out by multiplier ideal sheaves and the invariant whose non-vanishing obstructs
the existence of Kähler-Einstein metrics on Fano manifolds.

5.1 Introduction
One of the main problems in Kählerian and Sasakian geometries is the exis-
tence problem of Einstein metrics. An obvious necessary condition for the
existence of a Kähler-Einstein metric on a compact Kähler manifold M is that
the first Chern class c1 (M) is negative, zero or positive since the Ricci form
represents the first Chern class. This existence problem in Kählerian geometry
was settled by Aubin [1] and Yau [58] in the negative case and by Yau [58]
in the zero case. In the remaining case when the manifold has positive first
Chern class, in which case the manifold is called a Fano manifold in algebraic
geometry, there are two known obstructions. One is due to Matsushima [29]
which says that the Lie algebra h(M) of all holomorphic vector fields on a
compact Kähler-Einstein manifold M is reductive. The other one is due to
the first author [16] which is given by a Lie algebra character F : h(M) → C
with the property that if M admits a Kähler-Einstein metric then F vanishes
identically. Besides, it has been conjectured by Yau [59] that a more subtle con-
dition related to geometric invariant theory (GIT) should be equivalent to the
existence of Kähler-Einstein metrics. This idea was made explicit in the paper
[50] of Tian in which a notion called K-stability was introduced. Tian used a

68
 Multiplier ideal sheaves 69

generalized version of the invariant F for normal almost Fano varieties and
used it as the numerical invariant for the stability condition. The link between
the idea of GIT stability and geometric problems such as the existence problems
of Hermitian-Einstein metrics and constant scalar curvature Kähler metrics can
be explained through the moment maps in symplectic geometry. The explana-
tion from this viewpoint can be found for example in [14], [15], [13]. Recall
that an extremal Kähler metric is by definition a Kähler metric such that the
gradient vector field of the scalar curvature is a holomorphic vector field. In
particular, a Kähler metric of constant scalar curvature is an extremal Kähler
metric. The theorem of Matsushima is extended for extremal Kähler manifolds
by Calabi [3] as a structure theorem of the Lie algebra h(M) on an extremal
Kähler manifold M, and the first author’s obstruction F can be extended as
an obstruction to the existence of constant scalar curvature Kähler metric in a
fixed Kähler class ([17], [3]). The theorem of Calabi and the character F are
explained in the framework of the moment map picture by X. Wang [55] (see
also [19]).
In [10] Donaldson refined the notion of K-stability for a polarized manifold
(M, L), that is, a pair of an algebraic manifold M and an ample line bundle
L over M, and conjectured that there would exist a Kähler form in c1 (L)
of constant scalar curvature if and only if (M, L) is K-polystable. To define
K-(poly)stability for (M, L) Donaldson refined the invariant F even for non-
normal varieties which are degenerations of the polarized manifold (M, L) and
used it as the numerical invariant for the stability condition. The notion of
K-stability is defined as follows. For an ample line bundle L over a projective
variety M of dimension m, a test configuration of exponent r consists of the
following:

(1) A flat family of schemes π : M → C.

(2) A C∗ -action on M covering the usual C∗ -action on C.
(3) A C∗ -equivariant line bundle L → M such that
r for t = 0 one has Mt = π −1 (t) ∼
= M and (Mt , L|Mt ) ∼ = (M, Lr ),
m
r χ (Mt , Lt ) = p=0 (−1) dim H (Mt , Lt ) does not depend on t, in particu-
r p p r

lar for r sufficiently large dim H 0 (Mt , Lrt ) = dim H 0 (M, Lr ) for all t ∈ C.
Here we write Lrt for L|Mt though L may not exist for t = 0.

The C∗ -action on (L, M) induces a C∗ -action on the central fiber L0 → M0 =

π −1 (0). Moreover if (M, L) admits a C∗ -action, then one obtains a test config-
uration by taking the direct product M × C. This is called a product configu-
ration. A product configuration is called a trivial configuration if the action of
C∗ on M is trivial.
70 A. Futaki and Y. Sano

Definition 5.1.1 (M, L) is said to be K-semistable (resp. stable) if the invariant

F1 (defined below) of the central fiber (M0 , L0 ) is non-positive (resp. negative)
for all non-trivial test configurations. (M, L) is said to be K-polystable if it is
K-semistable and F1 = 0 only if the test configuration is product.
Here the invariant F1 is defined as follows. Let L0 → M0 be an ample line
bundle over an m-dimensional projective scheme. We assume that a C∗ -action
as bundle isomorphisms of L0 covers the C∗ -action on M0 . For any positive
integer k, there is an induced C∗ action on Wk = H 0 (M0 , Lk0 ). Put dk = dim Wk
and let wk be the weight of C∗ -action on ∧dk Wk . For large k, dk and wk are
polynomials in k of degree m and m + 1 respectively by the Riemann-Roch and
the equivariant Riemann-Roch theorems. For sufficiently large k we expand
wk
= F0 + F1 k −1 + F2 k −2 + · · · .
kdk
When M0 is smooth, F1 coincides with F (X) up to a negative multiple constant
where X is the infinitesimal generator of the C∗ -action on M0 .
The necessity of K-polystability for the existence of constant scalar curvature
Kähler metric has been studied by Chen and Tian [6], Paul and Tian [33], [34],
Donaldson [11], Stoppa [47] and Mabuchi [28].
Returning to Fano manifolds, there are two hopeful approaches to prove the
sufficiency of K-polystability for the existence of Kähler-Einstein metrics. One
is the Monge-Ampère equation and the other is the Kähler-Ricci flow. In both
cases the difficulty arises in the C 0 -estimate, and when the C 0 -estimate fails
the multiplier ideal sheaves and the subschemes cut out by them appear. The
multiplier ideal sheaves arising from the Monge-Ampère equation were studied
by Nadel [30], and those arising from Ricci flow were studied for example by
Phong, Sesum and Sturm [37] and Rubinstein [40]. We will give an overview
of this subject in section 2. On the other hand, for a given subscheme V in M,
Ross and Thomas [38] considered the test configuration obtained by blowing
up M × C along V × {0}. M is said to be slope stable if the invariant F1 for
the test configuration is negative for any V . These works lead us to ask how
the invariant F (or more generally F1 ) is related to the multiplier ideal sheaves
arising from the Monge-Ampère equation and the Ricci flow. We will treat this
subject in section 3.
Now we turn to Sasakian geometry. For general facts about Sasakian geom-
etry, refer to the book [2]. Let (S, g) be a Riemannian manifold. We denote its
Riemannian cone (R+ × S, dr 2 + r 2 g) by (C(S), ḡ). A Riemannian manifold
(S, g) is said to be a Sasaki manifold if the Riemannian cone (C(S), g) is
Kähler. From this definition the dimension of Sasaki manifold (S, g) is odd,
and we put dimR = 2m + 1 so that dimC C(S) = m + 1. (S, g) is isometric
 Multiplier ideal sheaves 71

to the submanifold {r = 1} = {1} × S ⊂ (C(S), g), and we identify S with the

submanifold {r = 1}. Let J be the complex structure on C(S) giving the Kähler
structure. Consider the vector field
∂

ξ = Jr .
∂r

2
 ξ to S ∼
Then (ξ − iJ ξ ) is a holomorphic vector field. The restriction ξ of 
1 
=
{r = 1} becomes a Killing vector field, called the Reeb vector field. The flow
generated by ξ is called the Reeb flow. The restriction η, of the 1-form  η on
C(S) defined as
1 √
η = 2 ḡ(ξ̃ , ·) = −1(∂¯ − ∂) ln r,

r
∼
to S = {r = 1} becomes a contact form. Hence dη defines Kähler forms on
local orbit spaces of Reeb flow. That is to say, the 1-dimensional foliation
defined by ξ comes equipped with a structure of transverse Kähler foliation.
A Sasaki manifold is said to be regular if the Reeb flow generates a free S 1 -
action, quasi-regular if all the orbits are closed. A Sasaki manifold is said to be
irregular if it is not quasi-regular.
For a polarized manifold (M, ω) the associated U (1)-bundle S of L becomes
a regular Sasaki manifold in a natural way: Choose a positive (1, 1) form
representing c1 (L), take the Hermitian metric h on L such that the connection
form η on L has its curvature form d η equal to ω. The Kähler cone C(S) is
L minus the zero section with the Kähler form given by 2i ∂∂r 2 where r is
the distance from the zero section. Conversely, any regular Sasaki manifold
is given in this way. Similarly, a quasi-regular Sasaki manifold is given as an
associated U (1)-orbibundle over an orbifold.
As is shown in [20] most of ideas in Kähler geometry can be extended to
transverse Kähler geometry for Sasaki manifolds. For example one can extend
Calabi’s theorem to compact Sasaki manifold whose transverse Kähler metric
is an extremal Kähler metric, and one can extend the obstruction F as an
obstruction for a basic cohomology class to admit a transverse Kähler form
with constant scalar curvature.
A Sasaki-Einstein manifold is a Sasaki manifold whose metric is an Ein-
stein metric. This condition is equivalent to the transverse Kähler metric being
Kähler-Einstein. Thus the study of the existence problem of Sasaki–Einstein
metrics are closely related to the problem of Kähler–Einstein metrics. But there
are differences between them. To explain the differences let k be the maximal
dimension of the torus which acts on C(S) by holomorphic isometries. When
k = m + 1 the cone C(S) is a toric variety, and in this case the Sasaki manifold
S is said to be toric. Notice that k is at least 1 because 
ξ generates holomorphic
72 A. Futaki and Y. Sano

isometries on C(S). The other extreme case is therefore when k = 1. In this

case the Sasaki manifold is necessarily quasi-regular.
The contact bundle D = Ker η ⊂ T S has a complex structure given by
the restriction of J . A necessary condition for the existence of Sasaki-Einstein
metric in a fixed transverse Kähler structure is that the following two conditions
are satisfied:
(a) the basic first Chern class is represented by a positive transverse (1, 1)-form,
(b) c1 (D) = 0,
see [2] or [20] for the proof. In [20] it is proved that if a compact toric Sasaki
manifold satisfies the conditions (a) and (b) then we can deform the Reeb vector
field so that the resulting Sasaki manifold has a Sasaki-Einstein metric. It can
be shown that the conditions (a) and (b) can be rephrased as the Sasaki manifold
is obtained from the toric diagram of constant height, and equivalently as the
apex of C(S) is a Q-Gorenstein singularity (c.f. [8]). In the case of the other
extreme when k = 1 the conditions (a) and (b) only say that the orbit space
of the Reeb flow is a Fano orbifold. In this case there is no deformation space
of Reeb vector field and the problem has the same difficulty as the problem of
Kähler-Einstein metrics. For the intermediate cases when 1 < k < m + 1 the
authors do not even know how to state the conjecture. In the extreme case where
k = 1 numerous existence results were obtained by Boyer, Galicki, Kollár and
their collaborators using the multiplier ideal sheaves, which will be reviewed
in the next section.

5.2 An overview of multiplier ideal sheaves

In this section, we recall the results about the relationships between the existence
of Kähler-Einstein metrics on Fano manifolds and multiplier ideal sheaves, and
related topics. In particular we focus on Nadel’s works and recent results about
multiplier ideal sheaves and the Kähler-Ricci flow.
Nadel [30] gave a sufficient condition for the existence of Kähler-Einstein
metrics on Fano manifolds by using multiplier ideal sheaves, which was
originally studied in the works of J.J. Kohn. Let M be a compact m-
dimensional Fano manifold. Let g be a Kähler metric on M, whose Kähler
class equals to the first Chern class c1 (M) of M. Let γ0 ∈ (0, 1). We denote
the Kähler form and the Ricci form of g by ωg and Ric(g) respectively.
Let
S := {ϕk ∈ CR∞ (M) | gi j¯ + ∂i ∂j¯ ϕk > 0, sup ϕk = 0, 1 ≤ k < ∞} (5.1)
M
 Multiplier ideal sheaves 73

be a sequence of Kähler potentials with respect to ωg such that


lim e−γ ϕk dV = ∞ (5.2)
k→∞ M

for any γ ∈ (γ0 , 1), and that there is a nonempty open subset U ⊂ M satisfying

e−ϕk dV ≤ O(1) (5.3)
U

as k → ∞, where dV is a fixed volume form. Note that the last condition (5.3)
always holds for any S due to gi j¯ + ∂i ∂j¯ ϕ > 0 (see for instance [49]). For each
S, Nadel constructed a coherent ideal sheaf I(S), which is called the multiplier
ideal sheaf (MIS). We will explain later the simpler definition of multiplier
ideal sheaves given by Demailly–Kollár [9].
Here let us recall the outline of Nadel’s construction. (See the original paper
[30] for the full details.) Let L be an arbitrary ample line bundle on M which
is not necessarily the anticanonical line bundle of M. We define H 0 (M, Lν )S
to be the set of all f ∈ H 0 (M, Lν ) for which there exists a sequence {fk } of
H 0 (M, Lν ) such that

|fk |2 e−γ ϕk dV ≤ C
M

for some γ ∈ (γ0 , 1) and fk → f uniformly. Consider the homogeneous coor-

dinate ring
∞
-
R(M, L) = H 0 (M, Lν ).
ν=0

Define
∞
-
I (M, S, L) = H 0 (M, Lν )S ,
ν=0

which is a homogeneous ideal I (M, S, L) of the graded ring R(M, L). Then,
the ideal sheaf I(S) is defined as the algebraic sheaf of ideals on M associated
to I (M, S, L). It is proved in [30] that this construction is independent of the
choice of L. Let V(S) be the (possibly non-reduced) subscheme in M cut out by
I(S). This subscheme is characterized as follows. A point p ∈ M is contained
in the complement of V(S) if and only if there exist an open neighborhood W
of p in M and a real number γ ∈ (γ0 , 1) such that

e−γ ϕk dV ≤ O(1)
W
74 A. Futaki and Y. Sano

as k → ∞. Note that V(S) is neither empty nor M if S satisfies the conditions

(5.1), (5.2) and (5.3).
One of the distinguished properties of this ideal sheaf is the following
vanishing theorem.
Theorem 5.2.1 [30] For every semi-positive Hermitian holomorphic line bun-
dle L on M,
H i (M, O(L) ⊗ I(S)) = 0, for all i > 0.
In particular, we have
H i (M, I(S)) = 0, i ≥ 0. (5.4)
Remark that (5.4) at i = 0 follows from the fact that the subscheme V(S) is not
empty. We also find that (5.4) implies
H 0 (V(S), OV(S) ) = C,
(5.5)
H i (V(S), OV(S) ) = 0 for all i > 0.
This vanishing formula (5.5) gives us several geometric properties of V(S). For
example,
(a) V(S) is connected.
(b) If V(S) is zero dimensional, then it is a single reduced point.
(c) If V(S) is one dimensional, then it is a tree of smooth rational curves.
The main result in [30] is that if a Fano manifold does not admit Kähler-Einstein
metrics then the bubble of the solution of the continuity method induces a proper
multiplier ideal sheaf with the above vanishing formula (5.5). To explain it, let
us recall the continuity method for the Monge-Ampère equation. Here, we
assume that γ0 = m/(m + 1). Consider the following equation
(det(gi j¯ ) + ∂i ∂j¯ ϕt )/(det(gi j¯ )) = exp(hg − tϕt ), (5.6)
where t ∈ [0, 1] and hg is the real-valued function defined by

√  
−1 ¯
Ric(g) − ωg = ∂ ∂hg , ehg ωgm = ωgm = V . (5.7)
2π M M

It is well-known that the space T := {t ∈ [0, 1] | (5.6) has a solution} contains

0 (due to the Calabi-Yau theorem) and open in [0, 1] (due to the implicit
√ function
−1 ¯
theorem). If T is closed then (5.6) is solvable at t = 1, i.e., ωg + 2π ∂ ∂ϕ1 gives
a Kähler-Einstein form. A priori estimates for the closedness of T were given
by Yau [58]; if (5.6) is solvable at s ∈ [0, t) and ϕs C 0 is uniformly bounded
then (5.6) is solvable at s = t. Nadel proved that if the solution {ϕt }0≤t<t0 of
 Multiplier ideal sheaves 75

(5.6) violates the above estimate, then there is a sequence {tk } such that tk → t0
as k → ∞ and {ϕtk − supM ϕtk }∞ k=1 induces a proper multiplier ideal sheaf I.
In this paper, we call it the Kähler-Einstein multiplier ideal sheaf (KE-MIS).
Summing up,

Theorem 5.2.2 ([30]) Let M be a Fano manifold which does not admit a
Kähler-Einstein metric. Let G be a compact subgroup of the group Aut(M)
of holomorphic automorphisms of M. Assume that M does not admit any
GC -invariant proper multiplier ideal sheaf. Then M admits Kähler-Einstein
metrics. Here GC denotes the complexification of G.

By combining the above theorem and the geometric properties of V(S) given
by the vanishing formula (5.5), Nadel gave many examples of Kähler-Einstein
Fano manifolds. Recently Heier [23] applied this method to (re-)prove the
existence of Kähler-Einstein metrics on complex del Pezzo surfaces obtained
from the blow up of CP2 at 3,4 or 5 points, which was originally proved by Siu
[44], Tian [49], Tian and Yau [51].
This method was extended to the case of Fano orbifolds by Demailly-Kollár
[9]. Their construction is simpler than [30]. Let ψ be an ωg -plurisubharmonic
(psh) function (or almost psh function with respect √
to ωg ), i.e., a real-valued
−1 ¯
upper semi-continuous function satisfying ωg + 2π ∂ ∂ψ ≥ 0 in the current
sense. The multiplier ideal sheaf with respect to ψ in the sense of [9] is the
ideal sheaf defined by the following presheaf

(U, I(ψ)) = {f ∈ O(U ) | |f |2 e−ψ dV < ∞} (5.8)
U

where U is an open subset of M. This sheaf is also coherent and satisfies the
vanishing theorem of Nadel type. In terms of this formulation, Theorem 5.2.2
can be written as follows. Let {ϕt } be the solution {ϕt }0≤t<t0 of (5.6) which
violates a priori estimates.

Theorem 5.2.3 ([9]) Let M be a Fano manifold of dimension m. Let G

be a compact subgroup of Aut(M). Assume that M does not admit a G-
invariant Kähler-Einstein metric. Let γ ∈ (m/(m + 1), 1). Then there exists
a G-invariant sequence {ϕtk }∞
k=1 such that

r tk → t0 as k → ∞,
r there exists a limit ϕ∞ = limk→∞ (ϕtk − supM ϕtk ) in the L1 -topology, which
is an ωg -psh function, and
r I(γ ϕ∞ ) is a GC -invariant proper multiplier ideal sheaf, i.e, I(γ ϕ∞ ) is
neither 0 nor OM .
76 A. Futaki and Y. Sano

We call I(γ ϕ∞ ) the KE-MIS of exponent γ . In the above, one of the

important ingredients is that the upper bound of γ is strictly smaller than 1.
To explain this point, we shall state Nadel’s vanishing theorem in terms of
Demailly-Kollár’s formulation. Let L be a holomorphic line bundle over M
with a singular Hermitian metric h = h0 e−ψ , where h0 is√a smooth Hermitian
−1 ¯
metric and ψ is a L1loc -function. Assume that h (L) = 2π ∂ ∂(− ln h0 + ψ)
is positive definite in the sense of currents, i.e., h (L) ≥ εωg for some ε > 0.
Then, in the same spirit of Nadel’s vanishing theorem, we have

H i (M, KM ⊗ L ⊗ I(ψ)) = 0, i > 0, (5.9)

where KM is the canonical line bundle. Now let ϕ be an ωg -psh function on a

−1
Fano manifold M with [ωg ] = c1 (M). Substitute KM and γ ϕ into L
√
and ψ in
−1 ¯
(5.9) respectively, and assume h0 associates to ωg . Since ωϕ = ωg + 2π ∂ ∂ϕ ≥
0, we have

h0 e−γ ϕ (L) = γ ωϕ + (1 − γ )ωg ≥ (1 − γ )ωg

if γ < 1. This means that the positivity condition for (5.9) with respect to
h0 e−γ ϕ holds if γ < 1. Then (5.9) implies

H i (M, I(γ ϕ)) = 0, i > 0.

Moreover, if the subscheme cut out by I(γ ϕ) is not empty, then

H 0 (M, I(γ ϕ)) = 0.

Summing up, we get

Lemma 5.2.4 If there exists a positive constant γ < 1 and an ωg -psh function
ϕ such that I(γ ϕ) is proper, then I(γ ϕ) satisfies (5.4). In particular, I(γ ϕ∞ )
for γ ∈ (m/(m + 1), 1) in Theorem 5.2.3 satisfies (5.4) (and then (5.5)).

On the other hand, the lower bound of γ in Theorem 5.2.3 describes the
strength of the singularity of ϕ∞ . It is closely related to a holomorphic invariant
introduced by Tian [49]. It is often called the α-invariant, which is defined by

αG (M) := sup{α ∈ R | e−α(ψ−supM ψ) ωgm < Cα for all G-invariant ωg -pshψ}
M
(5.10)
where G ⊂ Aut(M) is a compact subgroup. If a multiplier ideal sheaf I(γ ψ)
with respect to a G-invariant ωg -psh function ψ of exponent γ is proper,
where supM ψ = 0, then αG (M) ≤ γ , because e−γ ψ is not integrable over M.
Conversely,
 Multiplier ideal sheaves 77

Lemma 5.2.5 If αG (M) < 1, then there exist a positive constant γ ∈ (0, 1)
and a G-invariant ωg -psh function ψ with supM ψ = 0 such that I(γ ψ) is
proper.
Tian gave a sufficient condition for the existence of Kähler-Einstein metrics
on Fano manifolds in terms of this invariant.
Theorem 5.2.6 ([49]) If αG (M) > m/(m + 1), then M admits a G-invariant
Kähler-Einstein metric.
Using Theorem 5.2.6, Tian and Yau [51] proved the existence problem of
Kähler-Einstein metrics on Fano surfaces, i.e., the Fano surfaces obtained from
the blow up of CP2 at k points where 3 ≤ k ≤ 8 admits a Kähler-Einstein metric.
Both of the lower bound of αG (M) and the non-existence of the proper multiplier
ideal sheaves satisfying (5.5) give sufficient condition for the existence of
Kähler-Einstein metrics on Fano manifolds, and they are related directly to
each other. For example, Lemma 5.2.4 and 5.2.5, we have
Lemma 5.2.7 If αG (M) < 1, then a GC -invariant proper multiplier ideal
sheaves satisfying (5.5) exists.
Although αG (M) is difficult to compute in general, it is possible to calculate
it when M has a large symmetry in such cases as [46] for toric varieties and [12]
for the Mukai-Umemura 3-folds. On the other hand, there is a local version of
the αG (M)-invariant, which is called the complex singularity exponent [9].
Let K ⊂ M be a compact subset and ψ be a G-invariant ωg -psh function on
M. Then the complex singularity exponent cK (ψ) of ψ with respect to K is
defined by
cK (ψ) = sup{c ≥ 0 | e−cψ is L1 on a neighborhood of K}.
This constant depends only on the singularity of ψ near K. It is obvious
that cK (ψ) ≥ αG (M). One of the important properties of cK (ψ) is the semi-
continuity with respect ψ. Let P(M) be the set of all locally L1 ωg -psh functions
on M with L1 -topology. Then, we have (cf. Effective version of Main Theorem
0.2 in [9])
Theorem 5.2.8 ([9]) Let K ⊂ M be a compact subset of M. Let ϕ ∈ P(M)
be given. If c < cK (ϕ) and ψj → ϕ in P(M) as j → ∞, then e−cψj → e−cϕ
in L1 -norm over some neighborhood U of K.
In particular, if {ψj } satisfies

e−γ ψj dV → ∞
M
78 A. Futaki and Y. Sano

where γ ∈ (γ0 , 1) and ψj → ϕ in P(M), then cM (ϕ) ≤ γ0 . This theorem allows

us to substitute Theorem 5.2.3 for Theorem 5.2.2. In fact, if the solution ϕt of
(5.6) violates a priori C 0 -estimates at t = t0 , by using a Harnack inequality we
can show

e−γ (ϕt −sup ϕt ) dV → ∞ as t → t0
M

for any γ ∈ (m/(m + 1), 1). In Theorem 5.2.2, a subsequence of {ϕt } induces
the KE-MIS, which is proper. On the other hand, Theorem 5.2.8 implies that
e−γ ϕ∞ is not integrable over M for any γ ∈ (m/(m + 1), 1), where ϕ∞ :=
limi→∞ (ϕti − sup ϕti ). This means that ϕ∞ induces the KE-MIS in Theorem
5.2.3.
The multiplier ideal sheaves in [9] and the complex singularity exponent
can be defined algebraically as follows (cf. [27] and [2] for instance). Here we

consider a smooth variety M of dimension m. Let D = ai Di be a Q-divisor
on M. A log resolution of (M, D) is a projective birational map μ : M  → M
with M  smooth such that the divisor

μ∗ D + Ei
i

has simple normal crossing support. Assume D is effective and fix a log
resolution μ of (M, D). Then the multiplier ideal sheaf I(M, D) ⊂ OM with
respect to D is defined by

μ∗ OM  (KM  − μ∗ (!KM + D")), (5.11)

where !KM + D" means the integral part of KM + D. Note that I(M, D) is
independent of the choice of μ. This (algebraic) ideal sheaf corresponds to
the following (analytic) multiplier ideal sheaf defined in [9]. Take an open set
U ⊂ M so that for each Di there is a holomorphic function gi locally defining

Di in U . Let ϕD := i 2ai ln |gi | which is plurisubharmonic on U and define

|f |2
(U, I(ϕD )) := {f ∈ OM (U ) | . ∈ L1loc } (5.12)
|gi |2ai

as before. For simplicity, we assume that D = i ai Di has simple normal
crossing support. The holomorphic function f satisfies the L2 -integrability
.
condition in (5.12) if and only if f can be divided by g mi where mi ≥ !ai ",
i.e., I(ϕD ) = OM (−!D"). Let μ : M  → M be a log resolution of D. Then we
have

I(M, D) = μ∗ OM  (KM  − μ∗ (!KM + D")) = OM (−!D") = I(ϕD ).

 Multiplier ideal sheaves 79

The second equality in the above was proved in Lemma 9.2.19 [27]. We also
have

I(ϕD ) = OM ⇐⇒ e−ϕD ∈ L1 ⇐⇒ (M, D) is KLT. (5.13)

Here we say that a pair (M, D) is KLT if and only if

ordE (KM  − μ∗ (!KM + D")) > −1

for every exceptional divisor E with respect to a log resolution μ : M  → M. In

particular (M, D) is KLT is equivalent to that I(M, D) = OX . The equivalent
relation (5.13) essentially follows from that if Di is defined by {zi = 0} for
a local coordinate {zi } then the L1 -integrability of e−ϕD is equivalent to that
ai < 1 for all i (Proposition 3.20 [24]). In particular,

(M, γ D) is KLT ⇐⇒ e−γ ϕD ∈ L1 . (5.14)

Note that this holds for an (M, D) where D does not necessarily have simple
normal crossing support (Proposition 3.20 [24]). By using the KLT condition
(5.14), we can rephrase Theorem 5.2.3. Assume that a Fano manifold M does
not admit a Kähler-Einstein metric. Let ϕt be the solution of (5.6) where
t ∈ [0, t0 ). As explained before, by taking a subsequence of {ϕtj − supM ϕtj },
there exists a limit ϕ∞ in L1 -topology, which is an ωg -psh function, such
that e−γ ϕ∞ ∈ L1 for all γ ∈ (m/(m + 1), 1). Since an approximation theorem
in [9] implies that any ωg -psh function can be approximated by an ωg -psh

function formed of ln( i |fi |2 ) where all fi are holomorphic functions, we
can replace the above ϕ∞ by an ωg -psh function formed of 2s ln |τs | where
−s
τs ∈ H 0 (M, KM ) for sufficiently large s. That is to say, there exist a sufficiently
−s 2γ
) such that e−2γ s ln |τs | = |τs |− s ∈ L1 for
1
large integer s and τs ∈ H 0 (M, KM
−s
all γ ∈ (m/(m + 1), 1). Here | · | is the induced Hermitian metric on KM with
respect to the Kähler metric g. Hence we have the following theorem. Note that
the original result holds for orbifolds, but for simplicity we assume that M is
smooth in this paper.

Theorem 5.2.9 (Theorem 20 [24], Theorem 5.2.16 [2]) Let M be a Fano

manifold of dimension m and G be a finite subgroup of Aut(M). Assume that
m+ε
there is an ε > 0 such that a pair (M, m+1 D) is KLT for every G-invariant
−1
effective divisor D which is numerically equivalent to KM . Then M has a
G-invariant Kähler-Einstein metric.

The complex singularity exponent can be also defined algebraically as fol-

lows, which is called the log canonical thresholds (cf. Appendix in [5]). Let
Z ⊂ M be a closed subvariety. For an effective Q-Cartier divisor D on M, the
80 A. Futaki and Y. Sano

log canonical threshold of D along Z is defined by

lctZ (M, D) := sup{λ ∈ Q | the pair (M, λD) is log canonical along Z}.

Here, the pair (M, D) is called log canonical along Z if I(M, (1 − ε)D) is trivial
in a neighborhood of every point x ∈ Z for all 0 < ε < 1. For instance, let us
−l
consider a simple case. Let M be a Fano manifold and σ ∈ H 0 (M, KM ). Let ψσ
be an ω-psh function defined by ψσ (z) = 1l ln |σ (z)|. Let Dσ be the associated
divisor with σ and Z be a closed subvariety in M. In this case, lctZ (M, 1l D) is
the same as cZ (ψσ ). The log canonical threshold plays an important role in the
studies of the multiplier ideal (sheaves) in algebraic geometry (cf. [27]). Hence
we could expect that the complex singularity exponent with respect to the limit
ϕ∞ in Theorem 5.2.3 has something to do with the existence of Kähler-Einstein
metrics although it is not clear at the moment.
To find Kähler-Einstein metrics on Fano manifolds, there is another way
instead of solving (5.6), which is the (normalized) Kähler-Ricci flow. The
Ricci flow was introduced by R. Hamilton, and on a Fano manifold M with
Kähler class c1 (M) it is defined by
d
ωt = −Ric(ωt ) + ωt , ω0 = ωg (5.15)
dt
where t ∈ [0, ∞) and ωt is the Kähler form of the evolved Kähler metric gt .
Note that (5.15) is normalized so that the Kähler class of gt is preserved. The
existence and uniqueness of the solution of (5.15) for t ∈ [0, ∞) was proved
by Cao [4]. If (5.15) converges in C ∞ , the limit is a Kähler–Einstein metric.
Then, it is natural to ask whether or not the results about the multiplier ideal
sheaves obtained from the continuity method also hold for the Kähler–Ricci
flow. The first result of this issue was given by Phong–Sesum–Sturm [37] (see
also [36]). The equation (5.15) can be reduced to the equation at the potential
level
d
ϕt = ln(ωtm /ωgm ) + ϕt − hg , ϕ0 = c (5.16)
dt
√
−1 ¯
where c is a constant and ωt = ωg + 2π ∂ ∂ϕt . They gave a necessary and
sufficient condition condition for the convergence of ϕt as t → ∞. Their proof
consists of the parabolic analogue of Yau’s arguments for the elliptic Monge-
Ampère equation, the estimates about the Kähler-Ricci flow by Perelman
(cf. [43]) and the result about the Monge-Ampère equations by Kolodziej
([25], [26]).

Theorem 5.2.10 ([37]) For a certain appropriate constant c = c0 , the con-

vergence of the solution of (5.16) is equivalent to that there exists p > 1
 Multiplier ideal sheaves 81

such that

1
sup e−pϕt ωgm < ∞.
t≥0 V M

The convergence is then in C ∞ and exponentially fast.

To restate the above theorem in terms of the multiplier ideal sheaves, they
introduced the sheaf J p with respect to a family {ψt }0≤t<∞ of Kähler potentials,
which is defined by the presheaf

(U, J ) = {f ∈ O(U ) | sup
p
|f |2 e−pψt ωgm < ∞}. (5.17)
t≥0 M

Hence Theorem 5.2.10 implies the following.

Corollary 5.2.11 ([37]) The Kähler-Ricci flow converges if and only if there
exists p > 1 such that J p contains the global section 1.
The sheaf J p gives a necessary and sufficient condition for the existence
of Kähler-Einstein metrics and the lower bounds of p is optimal (cf. remarks
in [37]), whereas such results are not known for the case of the continuity
method. We emphasize that the sheaf J p contains different information from
the multiplier ideal sheaves in Theorem 5.2.3, because we do not need the limit
of {ϕt } but the whole of {ϕt } in order to define J p . In fact, in order to get the
limit of {ϕt }, we need the appropriate normalization of {ϕt } as Theorem 5.2.3. In
the terminology of the recent paper [45], the J p can be regarded as a dynamic
MIS which is similar to the Nadel’s formulation rather than a static MIS as the
Demailly-Kollár’s formulation. Now let us consider Theorem 5.2.10 by using
the static MIS instead of the dynamic MIS. Theorem 5.2.10 implies that if the
normalized Kähler-Ricci flow does not converge, then there is a subsequence
{ϕti }i of the solution of (5.16) such that

e−p(ϕti −sup ϕti ) ωgm → ∞
M

as i → ∞ for any p > 1. Hence, the limit ϕ∞ := lim(ϕti − sup ϕti ) implies
the multiplier ideal sheaf I(pϕ∞ ), which is proper for any p > 1. That is to
say, if there is no G-invariant ωg -psh function ψ such that I(pψ) is proper for
any p ∈ (1, +∞), then M admits a G-invariant Kähler-Einstein metric. More
precisely,
Theorem 5.2.12 [37] Let M be a Fano manifold. Let G ⊂ Aut(M) be a com-
pact subgroup. Assume that M does not admit Kähler-Einstein metrics. Let
p ∈ (1, ∞) and ωg ∈ c1 (M). There is a G-invariant subsequence of the solu-
tions {ϕkj }j ≥1 of (5.16) such that
82 A. Futaki and Y. Sano


r there exists the limit ϕ∞ = limj →∞ (ϕkj − 1 ϕ ωm ) in L1 -topology,
V M kj g
which is an ωg -psh function, and
r I(pϕ∞ ) is a GC -invariant proper multiplier ideal sheaf satisfying
−!p"
H i (M, I(pϕ∞ ) ⊗ KM ) = 0, for all i ≥ 1.

Note that Nadel’s vanishing formula (5.5) need not hold for the induced MIS
I(pϕ∞ ), because p > 1. However, this result still has an application. By using
a weaker version of Nadel’s vanishing theorem and Corollary 5.2.11, Heier [23]
proved the existence of Kähler-Einstein metrics for certain del Pezzo surfaces
with large automorphism group.
After [37], Rubinstein [40] gave an analogous result as Theorem 5.2.3 for
the Kähler-Ricci flow by using a static MIS as Demailly-Kollár. His proof is
similar to the case of the continuity method, and makes use of the estimates of
Perelman and the uniform Sobolev inequality of the Kähler-Ricci flow given
by Ye [60] and Zhang [61], which appeared after [37], in stead of Kolodziej’s
theorem.

Theorem 5.2.13 ([40]) Let M be a Fano manifold of dimension m. Let G

be a compact subgroup of Aut(M). Let γ ∈ (m/(m + 1), 1). Assume that M
does not admit a G-invariant Kähler-Einstein metric. Then there is an initial
constant c0 = ϕ0 and a G-invariant subsequence {ϕti } of the solution of (5.16)
such that

r there exists the limit ϕ∞ = limti →∞ (ϕti − 1 ϕ ωm ) in L1 -topology, which
V M ti g
is an ωg -psh function, and
r I(γ ϕ∞ ) is a GC -invariant proper multiplier ideal sheaf.

In this paper, we call the above multiplier ideal sheaf I(γ ϕ∞ ) the KRF
multiplier ideal sheaf of exponent γ . There are some remarks about the
above theorem. First, the KRF multiplier ideal sheaf is independent of the
choice of initial constant c0 of (5.16) due to the normalization of ϕti . In fact,
if we choose another constant c0 instead of the constant c0 in Theorem 5.2.13,
which is the same as Theorem 5.2.10, the solution of (5.16) is given by ϕt =
ϕt + (c0 − c0 )et . In contrast to this, when we consider the convergence of non-
normalized Kähler potentials {ϕt } as Theorem 5.2.10, we need to pay attention
to the choice of the constant c0 . Second, the normalization in Theorem 5.2.13
is equivalent to the one in Theorem  5.2.3. In fact, there is a uniform constant
C such that supM ϕt − C ≤ V1 M ϕt ωm ≤ supM ϕt . Third, γ is contained in
the interval (m/(m + 1), 1). This means that the subscheme cut out by I(γ ϕ∞ )
satisfies (5.5) and we can make use of the induced geometric properties. Fourth,
the process to prove Theorem 5.2.13 is similar to the case of the continuity
 Multiplier ideal sheaves 83

m
method, and the proof in [40] implies immediately that if αG (M) > m+1 then
the Kähler-Ricci flow will converge. (This similarity is pointed out in [7] after
[40] too.)
Rubinstein [41] also gave the analogous result of Theorem 5.2.12 in terms
of the discretization of the Kähler-Ricci flow called “Ricci iteration." Given a
Kähler form ω ∈ c1 (M) and a real number τ > 0, the time τ Ricci iteration is
defined by the sequence {ωkτ }k≥0 satisfying
ωkτ = ω(k−1)τ + τ ωkτ − τ Ric(ωkτ ) for k ∈ N, (5.18)
and ω0 = ω. When τ = 1, (5.18) is the discretization of (5.15). Let Hω be the
space of Kähler potentials with respect to (ω, c1 (M))
√
∞ −1 ¯
Hω := {ψ ∈ CR | ωψ = ω + ∂ ∂ψ > 0}.
2π
Let hωψ be the Ricci potential with respect to ωψ defined as (5.7). Since
[ωkτ ] = [ω(k−1)τ ], so (5.18) can be written as the system of complex Monge-
Ampère equations
ωψmkτ = ωm ehω + τ ϕkτ −ψkτ = ωψm(k−1)τ e( τ −1)ϕkτ − τ ϕ(k−1)τ ,
1 1 1
(5.19)
where k ∈ N, ωψkτ = ωkτ and ϕkτ := ψkτ − ψ(k−1)τ .
Theorem 5.2.14 [41] Let M be a Fano manifold. Let G ⊂ Aut(M) be a com-
pact subgroup. Assume that M does not admit Kähler-Einstein metrics. Let
τ = 1. Let γ ∈ (1, ∞) and ω ∈ c1 (M). There is a G-invariant subsequence of
the solutions {ψkj }j ≥1 of (5.19) such that

r there exists the limit ϕ∞ = limj →∞ (ψkj − 1 ψ ωm ) in L1 -topology,
V M kj
which is an ω-psh function, and
r I(γ ϕ∞ ) is a GC -invariant proper multiplier ideal sheaf satisfying
−!γ "
H i (M, I(γ ϕ∞ ) ⊗ KM ) = 0, for all i ≥ 1.
Considering Yau’s conjecture, it is also natural to ask how stability conditions
in the sense of GIT is related to the convergence of the Kähler-Ricci flow. For
example, see [35], [48] and [54] for references on this issue.

5.3 Direct relationships between multiplier ideal

sheaves and the obstruction F
It is conjectured by Yau that the existence of canonical Kähler metrics such
as Kähler-Einstein metrics and constant scalar curvature metrics for a given
84 A. Futaki and Y. Sano

Kähler class would be equivalent to the stability of manifolds in some sense

of Geometric Invariant Theory. This conjecture is formulated by Tian and
Donaldson in terms K-polystability as explained in section 1 and is still open.
This conjecture is an analogue of the so-called Hitchin-Kobayashi cor-
respondence, which was proved by Donaldson, and Uhlenbeck and Yau. The
proof of the direction from stability towards the existence of Hermitian-Einstein
metrics was preceded by constructing subsheaves which violate stability (such
sheaves are often called destabilizing subsheaves) from the bubble of the Yang-
Mills heat flow or the continuity method if a vector bundle does not admit a
Hermitian-Einstein metric. Weinkove [57] defined a MIS for each sequence of
Hermitian metrics on a holomorphic vector bundle and by using it he proved that
the bubble of the Yang-Mills heat flow induces a destabilizing subsheaf. Hence,
as the analogy between the Yau-Tian-Donaldson conjecture and the Hitchin-
Kobayashi correspondence, we could expect that the MIS obtained from the
continuity method or the Kähler-Ricci flow corresponds to a destabilizing sub-
sheaf in some sense for a Fano manifold with anticanonical polarization, but
their relation is not clear at this moment. This issue leads us to study direct
relationships between multiplier ideal sheaves and the obstruction F . Such a
direct relationship was first pointed out by Nadel in [31]. Extending Nadel’s
result is the main purpose of this section.
Up to this point we have not defined the character F explicitly, which we
do now. Let M be an m-dimensional Fano manifold with Kähler class c1 (M)
and g be a Kähler metric whose Kähler form ωg represents c1 (M). We denote
the Lie algebra consisting of all holomorphic vector fields on M by h(M). We
define the map F : h(M) → C by

F (v) := vhg ωgm .
M
In [16], the first author proved that F is independent of the choice of g and that
F is a Lie algebra character of h(M). If M is a Kähler-Einstein manifold, then
F vanishes on h(M) because we can take hg ≡ 0. Thus the vanishing of F is a
necessary condition for the existence of Kähler-Einstein metrics, but it is known
that it is not sufficient. For example, in [50] Tian gave a counterexample which
does not admit Kähler-Einstein metrics and have no nontrivial holomorphic
vector fields. So it is reasonable to study relationships between the invariant
F and multiplier ideal sheaves. First of all, we consider the multiplier ideal
sheaves obtained from the continuity method in the sense of Nadel. Assume
that M does not admit Kähler-Einstein metrics and that h(M) = {0}. For each
nontrivial holomorphic vector field v, define
Z + (v) := {p ∈ Zero(v) | Re(div(v)) > 0},
 Multiplier ideal sheaves 85

where Zero(v) is the zero set of v and div(v) is the divergence of v with respect to
some Kähler metric g, i.e., div(v) = (Lv (ωgm ))/ωgm , Lv being the Lie derivative
along v. Note that Z + (v) does not depend on the choice of g, although div(v)
does. Since M does not admit a Kähler-Einstein metric, the closedness of the
set of t  s for which the solutions {ϕt } of (5.6) exist does not hold, that is, the
solutions cease to exist at some t0 ∈ (0, 1]. Then the main result in [31] is as
follows.
Theorem 5.3.1 ([31]) Let M, h(M) and {ϕt } be as above. Let V be the induced
KE-MIS obtained from a subsequence of {ϕti }i where ti < t0 and ti → t0 . Then,
for any v ∈ h(M) with F (v) = 0, the support of V is not contained in Z + (v).

By using the above theorem, Nadel gave another theoretical approach to

show that CP1 does admit a Kähler-Einstein metric. In fact, if we assume that
CP1 did not admit a Kähler-Einstein metric, then V would be zero dimensional
and it would be a single reduced point, which follows from (b) of the properties
of the multiplier ideal subschemes. We may assume that V = {z = 0} in CP1 =
C ∪ {∞}. Let v = z dz d
∈ h(M), then v = 0 and the divergence of v is strictly
positive at z = 0. Hence V ⊂ Z + (v), which is a contradiction. As far as the
authors know, other applications of Theorem 5.3.1 except this example had
been unknown until [21].
We wish to extend this in several ways. We wish first of all to get some
more informations about Fano manifolds, secondly to show the existence of
MIS for Kähler-Ricci solitons, and thirdly to study the MIS arising from the
non-convergence of Kähler-Ricci flow and study the relation between MIS
and F .
We study three types of MIS: first of all KE-MIS which is due to Nadel, aris-
ing from the failure of solving Monge-Ampère equations for Kähler-Einstein
metrics by continuity method, secondly KRS-MIS which arises from the failure
of solving Monge-Ampère equations for Kähler-Ricci solitons by the continu-
ity method and thirdly KRF-MIS which arises from the failure of convergence
of Kähler-Ricci flow.
Let M be a Fano manifold, G a compact subgroup of Aut(M), and T r the
maximal torus of G. For any G-invariant Kähler metric g with
√
−1
ωg := g dzi ∧ dzj ∈ c1 (M)
2π ij
consider the Hamiltonian T r -action with the moment map μg : M → tr∗ . We
normalize it by

uX eh ωm = 0
M
86 A. Futaki and Y. Sano

where uX (p) = μ(p), X and Ricω − ω = i∂∂h. Note that this normalization
is equivalent to requiring uX to satisfy
uX + Xh + uX = 0,
see [18]. For ξ ∈ tr put
D ≤0 (ξ ) := {y ∈ μ(M) | < y, ξ > ≤ 0}.
Theorem 5.3.2 ([21]) Suppose M does not admit a Kähler-Einstein metric,
and let V be the support of the KE-MIS. Let ξ ∈ tr ⊂ h(M) satisfy F (vξ ) > 0
where vξ is the holomorphic vector field corresponding to ξ . Then
μg (V ) ⊂ D ≤0 (ξ )
for any G-invariant Kähler metric g whose Kähler form is in c1 (M).
Corollary 5.3.3 Let M be the one-point blow-up of CP2 . Then V is the excep-
tional divisor.
Note that this V destabilizes slope stability in the sense of Ross-Thomas by
a result of Panov and Ross [32].
Here is the outline of the proof of Theorem 5.3.2. Let h ∈ C ∞ (M) satisfy
Ricg − ωg = i∂∂h. Suppose
det(gij + ϕij )
= e−tϕ+h
det(gij )
has solutions only for t ∈ [0, t0 ), t0 < 1. Then we have an MIS with support
V . The following fact is due to Nadel based on earlier estimates by Siu and
Tian.
Fact 5.3.4 Let K ⊂ M − V be a compact subset of M − V . Then

ωgmt → 0
K
as t → t0 .
Fact 5.3.5
μg (p) ∈ D ≤0 (ξ ) ⇐⇒ (div(vξ ))(p) ≥ 0
where
div(vξ )(eh ωm ) = Lvξ (eh ωm ).
Fact 5.3.6

t
F (vξ ) = div(vξ )ωtm .
t −1 M
 Multiplier ideal sheaves 87

By Fact 5.3.6 and our assumption F (vξ ) > 0, we have for t ∈ (δ, t0 ) with
t0 < 1

t
div(vξ ) ωtm = F (vξ ) < −C
M t − 1
with C > 0 independent of t.
We seek a contradiction by assuming μg (V ) ⊂ D ≤0 (ξ ) = {div(vξ ) ≥ 0}.
Choose ! > 0 small and put

W! := {p ∈ M|div(vξ )(p) ≤ −!}.

Then W! ⊂ M − V is compact. Apply Fact 5.3.4 to W! to get


ωgmt → 0
W!

as t → t0 .
But then
  
−C ≥ div(vξ )ωtm = div(vξ )ωtm + div(vξ )ωtm
M M−W! W!
≥ −2!vol(M, g)

as t → t0 , a contradiction! This completes the outline of the proof of Theorem

5.3.2.
Next, we turn to KRS-MIS. Let M be again a Fano manifold of dimension
m. Let ωg ∈ c1 (M) be a Kähler form and v ∈ hr (M) a holomorphic vector field
in the reductive part hr (M) of h(M).

Definition 5.3.7 The pair (g, v) is said to be a Kähler-Ricci soliton if

Ric(ωg ) − ωg = Lv (ωg ).

(Hence Im(v) is necessarily a Killing vector field.)

Start with an initial metric g 0 with ω0 := ωg0 ∈ c1 (M).

 
Ric(ω0 ) − ω0 = i∂∂h0 , e ω0 =
h0 m
ω0m .
M M
 
iv ω0 = i∂θv,0 , eθv,0 ω0m = ω0m .
M M

Consider for t ∈ [0, 1]

 0   0  h0 −θv,0 −vϕt −tϕt
det gij + ϕtij = det gij e .
88 A. Futaki and Y. Sano

The solution for t = 1 gives the Kähler-Ricci soliton. Zhu [62] has shown that
t = 0 always has a solution. The implicit function theorem shows for some
! > 0, all t ∈ [0, !) have a solution.
Suppose we only have solutions on [0, t∞ ), t∞ < 1.
Let θv,g satisfy
 
iv ωg = i∂θv,g , e ωg =
θv,g m
ωgm .
M M

Definition 5.3.8 Define Fv : h(M) → C by


Fv (w) = w(hg − θv,g )eθv,g ωgm .
M

Tian and Zhu [52] showed that this Fv is independent of g with ωg ∈ c1 (M).

Theorem 5.3.9 (Tian-Zhu [52]) There exists a unique v ∈ hr (M) such that

Fv (w) = 0 for all w ∈ hr (M).

We take v to be the one chosen in the Theorem 5.3.9.

Theorem 5.3.10 ([21]) Let K be the compact subgroup such that k ⊗ C =

hr (M). Let v be the one chosen in the Theorem 5.3.9. Suppose there is no KRS.
Then we get an MIS and its support Vs satisfies

Vs ⊂ Z + (grad w) for ∀v ∈ hr (M).

Just as Nadel applied Theorem 5.3.1 to prove the existence of Kähler-

Einstein metric on CP1 , we can apply Theorem 5.3.10 to prove the existence
of KRS on the one point blow-up of CP2 .
Next we consider KRF-MIS. As mentioned in section 2, there are two
approaches to KRF-MIS, one by Phong, Sesum and Sturm [37], and the other
by Rubinstein [40]. Here, we consider the one considered by Rubinstein. So,
one gets an MIS from the failure of convergence of normalized Kähler-Ricci
flow:
∂g
= −Ric(g) + g. (5.20)
∂t
If we put gtij = gij + ϕtij the Ricci flow is equivalent to

∂ϕt det(gij + ϕtij )

= ln + ϕt − h0
∂t det(gij )
ϕ0 = c0
 Multiplier ideal sheaves 89

Rubinstein modified Phong–Sesum–Sturm’s MIS using the idea of Demailly–

Kollár:

ϕt − ϕt ωm −→ ϕ∞ almost psh
M

as t → ∞. Let Vγ be the MIS for ψ = γ ϕ∞ , γ ∈ ( m+1

m
, 1), defined by
  
(U, I(ψ)) = f ∈ OM (U ) | |f |2 e−ψ ωgm < ∞ .
U

This MIS satisfies

H q (M, I(ψ)) = 0 for ∀q > 0.
In general, it seems to be difficult to calculate (the support) of KRF-MIS.
However, under some assumptions, it becomes computable. To explain it, we
recall the following two results. Let M be a toric Fano manifold of dimension
m, on which the algebraic torus TC = (C∗ )m acts. Let TR = T m be the real torus
and tR its Lie algebra. We further put NR = J tR . Let W (M) = N (TC )/TC be
the Weyl group.
Theorem 5.3.11 (Wang-Zhu [56]) There exists a Kähler-Ricci Soliton
(gKRS , vKRS ).
Here we assume that K denotes a maximal compact subgroup of the reduc-
tive part of Aut(M) and KvKRS denotes the one-parameter subgroup of K gen-
erated by the imaginary part of vKRS . Then,
Theorem 5.3.12 (Tian-Zhu [53]) Let M be a (not necessarily toric) Fano
manifold which admits a Kähler-Ricci soliton (gKRS , vKRS ). Then, any solution
gt of (5.20) will converge to gKRS in the sense of Cheeger-Gromov if the initial
Kähler metric is KvKRS -invariant.
Combining Theorem 5.3.11 and Theorem 5.3.12, we find that the flow (5.20)
always converges to a Kähler-Ricci soliton in the sense of Cheeger-Gromov
on toric Fano manifolds. This fact suggests to us a possibility to understand
the asymptotic behavior of gt along (5.20) and to get some information about
KRF-MIS from data of Kähler-Ricci solitons. In fact, the second author proved
Theorem 5.3.13 ([42]) Suppose that the fixed point set NRW (M) of the Weyl
group W (M) on NR is one dimensional. Let σt = exp(tvKRS ) be the one
parameter group of transformations generated by vKRS , 0 < γ < 1 and ω a
TR -invariant Kähler form in c1 (M). Then the support of Rubinstein’s KRF-MIS
of exponent γ is equal to the support of the MIS of exponent γ obtained from
the Kähler potentials of {(σt−1 )∗ ω}.
90 A. Futaki and Y. Sano

Note that the assumption of NRW (M) is constrained and it would be expected
to be removed. Using the above theorem, the second author computed the
support of KRF-MIS for various γ in some examples. For example, we can
prove
Corollary 5.3.14 Let M be the blow up of CP2 at p1 and p2 . Let E1 and E2
be the exceptional divisors of the blow up, and E0 be the proper transform of
p1 p2 of the line passing through p1 and p2 . Then, the support of KRF-MIS on
M of exponent γ is
 2
∪i=0 Ei for γ ∈ ( 21 , 1),
E0 for γ ∈ ( 31 , 12 ).

It would be interesting to consider a relationship between destabilizing test

configurations and the pair of the support of KRF-MIS and its exponent.

References
[1] T. Aubin : Equations du type de Monge-Ampère sur les variétés kählériennes
compactes, C. R. Acad. Sci. Paris, 283, 119–121 (1976)
[2] C.P. Boyer and K. Galicki : Sasakian geometry, (Oxford Mathematical Mono-
graphs., 2008).
[3] E. Calabi : Extremal Kähler metrics II, Differential geometry and complex analysis,
(I. Chavel and H.M. Farkas eds.), 95–114, Springer-Verlag, Berline-Heidelberg-
New York, (1985).
[4] H.D. Cao : Deformation of Kähler metrics to Kähler-Einstein metrics on compact
Kähler manifolds, Invent. Math. 81 (1985) 359–372.
[5] I. Cheltsov and C. Shramov : Log canonical thresholds of smooth Fano threefolds
(with an appendix by J.P. Demailly), Uspekhi Mat. Nauk 63 (2008), no. 5(383),
73–180; translation in Russian Math. Surveys 63 (2008), no. 5, 859–958.
[6] X.X. Chen and G. Tian : Geometry of Kähler metrics and foliations by holo-
morphic discs, Publ. Math. Inst. Hautes Études Sci. No. 107 (2008), 1–107.
math.DG/0409433
[7] X.X. Chen and B. Wang : Remarks on Kähler Ricci flow, arXiv:0809.3963 (2008).
[8] K. Cho, A. Futaki and H. Ono : Uniqueness and examples of toric Sasaki-Einstein
manifolds, Comm. Math. Phys., 277 (2008), 439–458, math.DG/0701122
[9] J.P. Demailly and J. Kollár : Semi-continuity of complex singularity exponents
and Kähler-Einstein metrics on Fano orbifolds, Ann. Sci. École Norm. Sup. (4)
34, no.4 (2001) 525–556.
[10] S.K. Donaldson : Scalar curvature and stability of toric varieties, J. Differential
Geometry, 62(2002), 289–349.
[11] S.K. Donaldson : Lower bounds on the Calabi functional, J. Differential Geometry,
70(2005), 453–472.
 Multiplier ideal sheaves 91

[12] S.K. Donaldson : Kähler geometry on toric manifolds, and some other manifolds
with large symmetry, Handbook of geometric analysis. No. 1, 29–75, Adv. Lect.
Math. (ALM), 7, Int. Press, Somerville, MA, 2008. arXiv:0803.0985 (2008).
[13] S.K. Donaldson : Remarks on gauge theory, complex geometry and four-manifold
topology, in ‘Fields Medallists Lectures’ (Atiyah, Iagolnitzer eds.), World Scien-
tific, 1997, 384–403.
[14] S.K. Donaldson and P.B. Kronheimer : The geometry of four manifolds, Oxford
Mathematical Monographs, Claren Press, Oxford, 1990.
[15] A. Fujiki : Moduli space of polarized algebraic manifolds and Kähler metrics,
Sugaku Expositions, 5(1992), 173–191.
[16] A. Futaki : An obstruction to the existence of Einstein Kähler metrics, Invent.
Math. 73, 437–443 (1983).
[17] A. Futaki : On compact Kähler manifolds of constant scalar curvature, Proc. Japan
Acad., Ser. A, 59, 401–402 (1983).
[18] A. Futaki : Kähler-Einstein metrics and integral invariants, Lecture Notes in Math.,
vol.1314, Springer-Verlag, Berline-Heidelberg-New York, (1988).
[19] A. Futaki : Stability, integral invariants and canonical Kähler metrics, Proc. 9-th
Internat. Conf. on Differential Geometry and its Applications, 2004 Prague, (eds.
J. Bures et al), 2005, 45–58, MATFYZPRESS, Prague.
[20] A. Futaki, H. Ono and G. Wang : Transverse Kähler geometry of Sasaki manifolds
and toric Sasaki-Einstein manifolds, J. Differential Geom. 83 (2009), no. 3, 585–
635.
[21] A. Futaki and Y. Sano : Multiplier ideal sheaves and integral invariants on toric
Fano manifolds, arXiv:0711.0614 (2007).
[22] G. Heier : Convergence of the Kähler-Ricci flow and multiplier ideal sheaves on
Del Pezzo surfaces, Michigan Math. J. 58 (2009), no. 2, 423–440. arXiv:0710.5725
[23] G. Heier : Existence of Kähler-Einstein metrics and multiplier ideal sheaves on
Del Pezzo surfaces, Math. Z. 264 (2010), no. 4, 727–743. arXiv:0710.5724.
[24] J. Kollár : Singularities of pairs. (Algebraic Geometry–Santa Cruz 1995), Proc.
Sympos. Pure Math., 62, part 1 (1995) 221–287.
[25] S. Kolodziej : The complex Monge-Ampère equation, Acta Math. 180, no.1 (1998)
69–117.
[26] S. Kolodziej : The Monge-Ampère equation on compact Kähler manifolds, Indiana
Univ. Math. J. 52, no.3 (2003) 667–686.
[27] R. Lazarsfeld : Positivity in Algebraic Geometry II: Positivity for vector bundles,
and multiplier ideals, (Springer-Verlag, 2004).
[28] T. Mabuchi : K-stability of constant scalar curvature polarization, arXiv:0812.
4093
[29] Y. Matsushima : Sur la structure du groupe d’homéomorphismes d’une certaine
variété kaehlérienne, Nagoya Math. J., 11, 145-150 (1957).
[30] A.M. Nadel : Multiplier ideal sheaves and Kähler-Einstein metrics of positive
scalar curvature, Ann. of Math. (2) 132, no.3 (1990) 549–596.
[31] A.M. Nadel : Multiplier ideal sheaves and Futaki’s invariant, Geometric Theory of
Singular Phenomena in Partial Differential Equations (Cortona, 1995), Sympos.
Math., XXXVIII, Cambridge Univ. Press, Cambridge, 1998. (1995) 7–16.
[32] D. Panov and J. Ross : Slope Stability and Exceptional Divisors of High Genus,
Math. Ann. 343 (2009), no. 1, 79–101. arXiv:0710.4078v1 [math.AG].
92 A. Futaki and Y. Sano

[33] S.T. Paul and G. Tian : CM Stability and the Generalized Futaki Invariant I.,
math.AG/0605278, 2006.
[34] S.T. Paul and G. Tian : CM Stability and the Generalized Futaki Invariant II. (To
appear in Asterisque), math.AG/0606505, 2006.
[35] D.H. Phong and J. Sturm : On stability and the convergence of the Kähler-Ricci
flow, J. Differential Geom. 72 (2006), no. 1, 149–168.
[36] D.H. Phong and J. Sturm : Lectures on stability and constant scalar curvature,
Handbook of geometric analysis, No. 3, 357–436, Adv. Lect. Math. (ALM), 14,
Int. Press, Somerville, MA, 2010. arXiv:0801.4179.
[37] D.H. Phong, N. Sesum and J. Sturm : Multiplier ideal sheaves and the Kähler-Ricci
flow, Comm. Anal. Geom. 15, no. 3 (2007), 613–632.
[38] J. Ross and R.P. Thomas : An obstruction to the existence of constant scalar
curvature Kähler metrics, J. Differential Geom. 72 (2006), no. 3, 429–466.
[39] Y.A. Rubinstein : The Ricci iteration and its applications, C. R. Acad. Sci. Paris,
Ser. I, 345 (2007), 445-448, arXiv:0706.2777.
[40] Y.A. Rubinstein : On the construction of Nadel multiplier ideal sheaves and the
limiting behavior of the Ricci flow, Trans. Amer. Math. Soc. 361 (2009), no. 11,
5839–5850. arXive:math/0708.1590.
[41] Y.A. Rubinstein : Some discretizations of geometric evolution equations and the
Ricci iteration on the space of Kähler metrics, Adv. Math. 218 (2008), no. 5,
1526–1565. arXive:math/0709.0990.
[42] Y. Sano : Multiplier ideal sheaves and the Kähler-Ricci flow on toric Fano mani-
folds with large symmetry, preprint, arXiv:0811.1455.
[43] N. Sesum and G. Tian : Bounding scalar curvature and diameter along the Kähler-
Ricci flow (after Perelman), J. Inst. Math. Jussieu 7 (2008), no. 3, 575–587.
[44] Y-T. Siu : The existence of Kähler-Einstein metrics on manifolds with positive
anticanonical line bundle and a suitable finite symmetry group, Annals of Math.
127 (1988) 585–627.
[45] Y-T. Siu : Dynamical multiplier ideal sheaves and the construction of rational
curves in Fano manifolds, Complex analysis and digital geometry, 323–360, Acta
Univ. Upsaliensis Skr. Uppsala Univ. C Organ. Hist., 86, Uppsala Universitet,
Uppsala, 2009. arXiv:0902.2809.
[46] J. Song : The α-invariant on toric Fano manifolds, Amer. J. Math. 127, No.6 (2005)
1247–1259.
[47] J. Stoppa : K-stability of constant scalar curvature Kähler manifolds, Adv. Math.
221 (2009), no. 4, 1397–1408. arXiv:0803.4095.
[48] G. Székelyhidi : The Kähler-Ricci flow and K-polystability, Amer. J. Math. 132
(2010), no. 4, 1077–1090. arXiv:0803.1613.
[49] G. Tian : On Kähler-Einstein metrics on certain Kähler manifolds with C1 (M) > 0,
Invent. Math. 89, no.2 (1987) 225–246.
[50] G. Tian, Kähler-Einstein metrics with positive scalar curvature, Invent. Math. 130,
no.1 (1997) 1–37.
[51] G. Tian and S.T. Yau, Kähler-Einstein metrics on complex surfaces with C1 (M)
positive, Comm. Math. Phys. 112 (1987) 175–203.
[52] G. Tian and X. Zhu : A new holomorphic invariant and uniqueness of Kähler-Ricci
solitons, Comment. Math. Helv 77, No.2 (2002) 297–325.
 Multiplier ideal sheaves 93

[53] G. Tian and X. Zhu : Convergence of Kähler-Ricci flow, J. Amer. Math. Soc., 20,
No.3 (2007), 675–699.
[54] V. Tosatti : Kähler-Ricci flow on stable Fano manifolds, J. Reine Angew. Math.
640 (2010), 67–84. arXiv:0810.1895.
[55] X.-W. Wang : Moment maps, Futaki invariant and stability of projective manifolds,
Comm. Anal. Geom. 12 (2004), no. 5, 1009–1037.
[56] X.J. Wang and X. Zhu : Kähler-Ricci solitons on toric manifolds with positive first
Chern class, Adv. Math. 188, No.1 (2004) 87–103.
[57] B. Weinkove, A complex Frobenius theorem, multiplier ideal sheaves and
Hermitian-Einstein metrics on stable bundles, Trans. Amer. Math. Soc. 359, no.4
(2007) 1577–1592.
[58] S.-T.Yau : On the Ricci curvature of a compact Kähler manifold and the complex
Monge-Ampère equation I, Comm. Pure Appl. Math. 31(1978), 339–441.
[59] S.-T. Yau : Open problems in Geometry, Proc. Symp. Pure Math. 54 (1993) 1–28.
[60] R. Ye : The logarithmic Sobolev inequality along the Ricci flow, arXiv:0707.2424
(2007).
[61] O.S. Zhang : A uniform Sobolev inequality under Ricci flow, arXiv:0706.1594
(2007).
[62] X. Zhu, Kähler-Ricci soliton type equations on compact complex manifolds with
C1 (M) > 0, J. Geom. Anal., 10 (2000), 759–774.

Authors’ addresses:

Akito Futaki
Department of Mathematics,
Tokyo Institute of Technology, 2-12-1,
O-okayama, Meguro, Tokyo 152-8551,
Japan
[email protected]

Yuji Sano
Department of Mathematics,
Kyushu University,
6-10-1, Hakozaki, Higashiku,
Fukuoka-city, Fukuoka 812-8581
Japan
[email protected]
 6
Multisymplectic formalism and the covariant
phase space
fr éd éric h élein

In most attempts to build the mathematical foundations of Quantum Field

Theory (QFT), two classical ways have been explored. The first one is often
referred to as the Feynman integral or functional integral method. It is a gen-
eralization to fields of the path integral method of quantum mechanics and is
heuristically based on computing integrals over the infinite dimensional set of
all possible fields ϕ by using a kind of ‘measure’ – which should behave like the
Lebesgue measure on the set of all possible fields ϕ – times eiL(ϕ)/ , where L
is a Lagrangian functional (but attempts to define this ‘measure’ failed in most
cases). The second one is referred to as the canonical quantization method and
is based on the Hamiltonian formulation of the dynamics of classical fields, by
following general axioms which were first proposed by Dirac and later refined.
The Feynman approach has the advantage of being manifestly relativistic, i.e.
it does not require the choice of a particular system of space-time coordinate,
since the main ingredient is L(ϕ), which is an integral over all space-time. By
contrast, the canonical approach, at least its classical formulation, seems to be
based on the choice of a particular time coordinate which is needed to define
the Hamiltonian function through an infinite dimensional Legendre transform.
However there are alternative formulations of the Hamiltonian structure of
the dynamics of classical fields, which could be used as a starting point of a
covariant canonical quantization1 . We shall see two of them in this text: the
covariant phase space and the multisymplectic formalism. The covariant phase
space is based on the observation that the set of classical solutions to a vari-
ational problem, i.e. of critical points of some action functional L(ϕ), has an
intrinsic canonical symplectic (or presymplectic) structure. The multisymplec-
tic formalism is a generalization of the standard symplectic formalism, where
1 The word ‘covariant’ refers here to a construction which does not rely on the choice of a
particular system of coordinates on space-time and hence which respects the basic principles of
Relativity.

94
 Multisymplectic formalism and the covariant phase space 95

the time of classical mechanics is replaced  by space-time: for instance if we

start from a Lagrangian action L(ϕ) = X L(x, ϕ, dϕ) we do not perform a
Legendre transform with respect to a chosen time coordinate, but with respect
to all space-time coordinates.
We expect that, roughly speaking, both the Feynman and the canonical
approach should lead to equivalent theories. However if this fact is true, it
should not be trivial for several reasons. A first obvious remark is that both
theories are only heuristic and have no mathematical foundations, except in very
simplified situations. A superficial difference between both approaches is the
fact that one is based on the Lagrangian, the other one the Hamiltonian function.
Moreover these two approaches answer different questions: Feynman’s offers
a short and intuitive way to compute the quantities which can be measured in
interaction processes between particles (although one cannot avoid the difficult
step of regularizing and renormalizing the computed quantities). For the same
task, the canonical approach seems to be more complicated; however it proposes
a scheme to build mathematical objects (a complex Hilbert space of physical
states and an algebra of self-adjoint operators acting on it, corresponding to
observable quantities), the construction of which requires more effort by using
the Feynman integral. But a deep difference between the methods is that the
Feynman integral is a construction off shell, i.e. on the set of all possible
fields, even those which are not solutions of the classical dynamical equations,
whereas in many cases the canonical approach is a construction on shell, i.e. on
the set of fields which are solutions of the dynamical equations (in particular
in the covariant phase space method).
In this paper we shall present briefly the multisymplectic formalism and the
covariant phase space and show the strong relation between both theories. To
my knowledge this relation was discovered by J. Kijowski and W. Szczyrba in
1976 [25], but their beautiful paper seems to have been ignored in the literature.
We have included some historical comments. We shall conclude by presenting
the geometric quantization scheme for linear field equations (i.e. free fields
in the terminology of physicists) in the framework of multisymplectic geometry.
The goal is to show how canonical quantization can be performed in a covariant
way.

6.1 The multisymplectic formalism

6.1.1 Maps between vector spaces
We start with a simple variational problem: let X and Y be two vector space
of dimension n and k respectively and assume that X is oriented, let U be an
96 F. Hélein

open subset of X and consider the set C ∞ (U, Y) of smooth maps u from U to
Y. Let L : U × Y × End(X, Y) −→ R be a Lagrangian density and consider
the action functional on C ∞ (U, Y) defined by:

L[u] = L(x, u(x), dux )β,
U

where β is a volume n-form on U . We use coordinates (x 1 , . . . , x n ) on U

s.t. β = dx 1 ∧ · · · ∧ dx n , coordinates (y 1 , . . . , y k ) on Y and vμi on End(X, Y).
Then the critical points of L satisfy the Euler–Lagrange system of equations
!
∂ ∂L ∂L
μ i
(x, u(x), dux ) = i (x, u(x), dux ), ∀i s.t. 1 ≤ i ≤ k. (6.1)
∂x ∂vμ ∂y

We assume that the map

U × Y × End(X, Y) −→ U
 × Y∂L × End(Y∗, X∗ )
(x, y, v) −→ x, y, ∂v (x, y, v) ,
is a diffeomorphism: this is the analogue of the Legendre hypothesis used in
Mechanics. We denote by p∗ = (pi )μ,i the coordinates on End(Y∗ , X∗ ) and
μ

we define the Hamiltonian function H on U × Y × End(Y∗ , X∗ ) by

H (x, y, p∗ ) := pi vμi − L(x, y, v),

μ

where we assume implicitly that v = (vμi )i,μ is the unique solution of

∂L
∂v
(x, y, v) = p∗ . Then vμi is actually equal to ∂p ∂H
μ . Moreover to any map
i
u : U −→ Y we associate the map p : U −→ End(Y∗ , X∗ ) s.t. p∗ (x) :=
∗
∂L
∂v
(x, u(x), dux ), ∀x ∈ U . Then we can show [45] that u is a solution of
(6.1) iff (u, p∗ ) is a solution of the generalized Hamilton system:
⎧ i
⎪
⎪ ∂u ∂H ∗
⎨ ∂x μ (x) = ∂pμ (x, u(x), p (x))
⎪
i
(6.2)
⎪
⎪ ∂p μ
∂H
⎪
⎩ μ (x) = − i (x, u(x), p (x)).
i ∗
∂x ∂y
System (6.2) can be translated as a geometric condition [23] on the graph

 ∗ := G(u, p∗ ) := {(x, u(x), p∗ (x))| x ∈ U } ⊂ U × Y × End(Y∗ , X∗ ).

Indeed consider a family of n vector fields X1 , . . . , Xn : U −→ U × Y ×

End(Y∗ , X∗ ) s.t. for any x ∈ U , (X1 (x), . . . , Xn (x)) is a basis of the tangent
plane to G(u, p∗ ) at (x, u(x), p∗ (x)). Set βμ := ∂x∂ μ β. Then (6.2) is equiva-
lent to the condition that ∀ξ ∈ X × Y × End(Y∗ , X∗ ),
μ
dpi ∧ dy i ∧ βμ (ξ, X1 , . . . , Xn ) = dH (ξ )β(X1 , . . . , Xn ). (6.3)
 Multisymplectic formalism and the covariant phase space 97

μ
i ∂p
In fact this can be easily checked by choosing Xμ := ∂x∂ μ + ∂x μ ∂y i + ∂x μ ∂p μ .
∂u ∂ i ∂
i
More concisely we can introduce the n-multivector field X := X1 ∧ · · · ∧ Xn
(so that X(x) ∈ "n T(x,u(x),p∗ (x))  ∗ , ∀x ∈ U ). Then Equation (6.3) reads:

∀ξ ∈ X × Y × End(Y∗ , X∗ ),
μ
dpi ∧ dy i ∧ βμ (ξ, X) = dH (ξ )β(X). (6.4)

Equation (6.3) can be completed with the independence condition

β|∗ = 0, (6.5)

where, if j∗ :  ∗ −→ U × Y × End(Y∗ , X∗ ) denotes the inclusion map,

β|∗ := j∗∗ β. This condition guarantees that locally  ∗ is the graph of some
map (u, p∗ ) over the ‘space-time’ X.
We will see now that the independence condition (6.5) can be further incor-
porated in a dynamical condition analogous to (6.3) by adding to the variables
(x, y, p∗ ) a variable e dual to β. We define M := U × Y × R × End(Y∗ , X∗ )
with coordinates (x, y, e, p∗ ) = (x μ , y i , e, pi ) and the (n + 1)-form
μ

ω := de ∧ β + dpiμ ∧ dy i ∧ βμ . (6.6)

We define H : M −→ R by H(x, y, e, p∗ ) := e + H (x, y, p∗ ). Then to any

oriented n-dimensional submanifold  ∗ = G(u, p∗ ) we associate the oriented
n-dimensional submanifold  := {(x, u(x), e(x), p∗ (x))| x ∈ U } of M, where
e is such that e(x) + H (x, u(x), p∗ (x)) = h, ∀x ∈ X, for some real constant2
h. Then  ∗ is a solution of (6.3) and (6.5) iff  is a solution of:

∀ξ ∈ C ∞ (M, T M), ω(ξ, X) = dH(ξ )β(X), (6.7)

where C ∞ (M, T M) denotes the set of sections of T M over M, which can be

identified with C ∞ (M, X × Y × R × End(Y∗ , X∗ )). Note that conversely it is
easy to check that any connected solution  of (6.7) is contained in a level set of
H. We call a Hamiltonian n-curve any solution  of (6.7). The (n + 1)-form
ω is an example of a multisymplectic form and the pair (M, ω) is called a
multisymplectic manifold. Using the notation ξ ω for the interior product of
the vector ξ with the (n + 1)-form ω, we set:

Definition 6.1.1 Let M be a smooth manifold. A multisymplectic (n + 1)-

form ω on M is a (n + 1)-form which is closed (i.e. dω = 0) and which is
non degenerate (i.e. ∀m ∈ M, ∀ξ ∈ Tm M, ξ ω = 0 =⇒ ξ = 0).

2 W.l.g. we can assume that the constant h is zero, so that  is included in H−1 (0).
98 F. Hélein

6.1.2 Higher order problems

The precedingcan be extended to an action on maps u : U −→ Y of the
form L[u] := U L(x, j r u(x))β, where j r u denotes the r-th order jet of u
(i.e. all partial derivatives of u of order less than or equal to r). We denote
by J r (U, Y) the r-th order jet space of maps from U to Y and we use the
coordinates x = (x μ )μ and v = vμi 1 ···μa i,μ ···μ (for 1 ≤ i ≤ k, 0 ≤ a ≤ r and
1 a
1 ≤ μb ≤ n) on J r (U, Y) s.t.

  ∂ a ui
vμi 1 ···μa j r u(x) = (x).
∂x μ1 · · · ∂x μa
It is convenient to introduce the multi-index notation M = μ1 · · · μa , where
a ∈ N and ∀b ∈ [[1, a]], 1 ≤ μb ≤ n and to set |M| = a. Then for |M| = r we
denote
∂L
πiM (x, v) := i
(x, v). (6.8)
∂vM

The analogue of the Legendre hypothesis consists here in supposing that the
map (vM i
)i,M;|M|=r −→ (piM )i,M;|M|=r defined by (6.8) is one to one. Next we
define the vector space M with coordinates:

(x, v) : xμ vi vμi ··· vμi 1 ···μr−1

μ ···μ μ ···μr
p = (e, p∗ )
μ
: e pi ··· pi 1 r−1 pi 1

for 1 ≤ μ, μb ≤ n and 1 ≤ i ≤ k. Clearly M contains J r−1 (U, Y) as a vector

subspace. We also define recursively, for |M| ≤ r − 1,
∂L Mμ
πiM (x, v) := i
(x, v) − Dμ πi (x, vM ),
∂vM

where
∂ ∂
Dμ := μ
+ vMμ
i
i
.
∂x ∂vM

Then we define a Hamiltonian function on M:

μ ···μr i
H (x, v, p∗ ) := pi vμi + · · · + pi 1
μ
vμ1 ···μr − L(x, v),

where we assume implicitly that, for |M| = r, vM i

= vμi 1 ···μr is the solution
of pi = πi (x, v), ∀M s.t. |M| = r, and we set piM := πiM (x, v), ∀M s.t.
M M

|M| < r. To any map u from U to Y we associate the map p∗ which is the
image of j r u by the maps πiM . Then u is a critical point of L iff (j r−1 u, p∗ ) is
 Multisymplectic formalism and the covariant phase space 99

a solution of the generalized Hamilton equations [8]

⎧
⎪ ∂uiμ1 ···μa ∂H
⎪
⎨ (x) = ∗
μ ···μ μ (x, u(x), p (x)) for 0 ≤ a ≤ r − 1
∂x μ ∂pi 1 a
μ ···μ μ (6.9)
⎪
⎪ ∂p 1 a ∂H
⎩ i μ (x) = − i (x, u(x), p∗ (x)) for 0 ≤ a ≤ r − 1,
∂x ∂vμ1 ···μa
Alternatively we can consider the map x −→ (x, j r−1 u(x), e(x), p∗ (x)), where
e may be chosen so that e(x) + H (x, j r−1 u(x), p∗ (x)) = 0, ∀x and we can write
(6.9) in a way similar to (6.7) by using the Hamiltonian function
H(x, v, p) = e + H (x, v, p∗ ).
and the multisymplectic form
μ μ ···μr
ω := de ∧ β + dpi ∧ dv i ∧ βμ + · · · + dpi 1 ∧ dvμi 1 ···μr−1 ∧ βμr .
An intrinsic geometrical multisymplectic formulation of these equations has
been derived recently by L. Vitagliano [43].

6.1.3 More general multisymplectic manifolds

Assume that we start with an action L which is an integral of a Lagrangian
density which depends on the first order derivatives of the field. This may be for
instance a variational problem on maps u between two manifolds X and Y with
a functional L[u] := X L(x, u(x), dux )β or a variational problem on sections
of a fiber bundle π : Z −→ X . Then a natural multisymplectic manifold is the
vector bundle "n T ∗ (X × Y) in the first case or "n T ∗ Z in the second case.
Both manifolds are indeed endowed with a canonical (n + 1)-form ω which is
the straightforward analogue of the canonical symplectic form on any cotangent
bundle [6, 24, 19]. We may call this manifold the universal multisymplectic
manifold associated with the Lagrangian problem. Although this construction
seems to be similar to the symplectic one for Hamiltonian mechanics, it is
different because, say for maps between two manifolds X and Y of dimensions
n and k respectively and a first order variational problem, on the one hand
the Lagrangian density depends on n + k + nk variables (in other words the
analogue of the product of the time real line and of the tangent bundle in
mechanics has dimension n + k + nk), whereas on the other hand the analogue
of the cotangent bundle is "n T ∗ (X × Y) and has dimension n + k + (n+k)! n!k!
.
This means that we have much more choices in the Legendre transform, which
is not a map in general but a correspondence, as soon as n ≥ 2 and k ≥ 2.
This is why it is possible and often simpler to impose extra constraints in the
Legendre transform, which means that we replace the universal multisymplectic
100 F. Hélein

manifold "n T ∗ Z (whatever Z is: a Cartesian product X × Y or the total

space of a bundle) by some submanifold of it. Most authors prefer to use the
affine multisymplectic submanifold "n2 T ∗ Z: if Z = X × Y, it is the subbundle
of "n T ∗ (X × Y) over X × Y, the fiber over the point (x, y) ∈ X × Y of
∗
which is the subspace of n-forms p ∈ "n T(x,y) (X × Y) s.t. ∀η1 , η2 ∈ Ty Y,
(0, η1 ) ∧ (0, η2 ) p = 0. If Z is a fiber bundle over X , "n2 T ∗ Z, which is the
subbundle over Z, the fiber over z ∈ Z of which is the subspace of n-forms
p ∈ "n Tz∗ Z s.t. for any pair of vertical vectors η1 , η2 ∈ Tz Z, η1 ∧ η2 p = 0
(by ‘vertical’ we mean that η1 and η2 are maps to 0 ∈ Tπ(z) X by the differential
μ
of π ). In both cases an n-form reads p = eβ + pi dy i ∧ βμ in local coordinates
and the latter theory is actually the right generalization of (6.6). This theory is
usually referred to as the De Donder–Weyl theory although it was discovered
by V. Volterra (see §6.1.8). Note that "n2 T ∗ Z can alternatively be defined as
being the affine dual of the first jet bundle of sections of Z over X [14].

6.1.4 Premultisymplectic manifolds

A variant consists in manifolds equipped with a closed (n + 1)-form but without
assuming a non-degeneracy condition, as for instance in [25]:
Definition 6.1.2 We call a triple (M, ω, β) an n-phase space if M is a mani-
fold, ω is a closed (n + 1)-form, called a premultisymplectic form and β is a
non vanishing n-form.
Examples of premultisymplectic manifolds can be built easily by starting
from a multisymplectic manifold (M, ω) with a Hamiltonian function H on it
which has no critical points (as for instance H(x, y, e, p∗ ) = e + H (x, y, p∗ )
for the previous theory). Then we let η be a vector field on M s.t. dH(η) = 1
everywhere and we set β := η ω. For any h ∈ R the level set Mh := H−1 (h)
is a submanifold. Then (Mh , ω|Mh , β|Mh ) is a premultisymplectic manifold
[18]. In particular ω|Mh is obviously closed but may be degenerate in general:
indeed if  is a Hamiltonian n-curve contained in Mh then any vector tangent
to  is in the kernel of ξ −→ ξ ω. In fact an n-phase space (M, ω) carries
an intrinsic dynamical structure: we say that an n-dimensional submanifold 
of M is a Hamiltonian n-curve if:
∀v ∈ C ∞ (M, Tm M), (v ω) | = 0 and β| = 0. (6.10)
This definition is motivated by the fact that (if  is connected)  is a solution of
(6.7) iff there exists some h ∈ R s.t.  is contained in Mh and  is a Hamiltonian
n-curve in the n-phase space (Mh , ω|Mh , β|Mh ) (see [18]). However there
are examples of premultisymplectic manifolds which do no arise from this
 Multisymplectic formalism and the covariant phase space 101

construction as for instance the example in [20, 35] obtained by starting from
the Palatini formulation of gravity.

6.1.5 Action principle

We assume here that we are working in a premultisymplectic manifold
(M, ω, β) and that the form ω is exact, i.e. is of the form ω = dθ , where
θ is an n-form on M. This is true e.g. in a submanifold of "n T ∗ Z, where ω is
precisely defined as the differential of a canonical ‘Poincaré–Cartan’ form θ .
To any oriented n-dimensional submanifold  we associate the action

A[] := θ. (6.11)


One can then show that any n-dimensional submanifold  on which β does
not vanish is a critical point of A iff it is a Hamiltonian n-curve, i.e. a solution
of (6.10) (see [18]). Actually if  is the image of a given configuration by
some Legendre transform, then A[] coincides with the Lagrangian action of
the configuration we started with [19]. Note that in the case where ω is not
exact one could define a similar
 action
 on a homology class of n-dimensional
submanifolds by replacing  by  ω, where  is a (n + 1)-chain connecting
 with a particular n-dimensional submanifold which generates the homology
class.

6.1.6 Observable functionals

An observable functional is a functional on the ‘space’ of all solutions: this
notion will be central in the next section concerning the covariant phase space.
A particular class of such functionals arise in the context of multisymplectic
manifolds or premultisymplectic manifolds as follows. In the following we
denote by F the set of n-dimensional oriented submanifolds (fields) of M and
by E the subset of F composed of Hamiltonian n-curves.

In a multisymplectic manifold (M, ω)

We define an infinitesimal symplectomorphism of (M, ω) to be a vector field
ξ ∈ C ∞ (M, T M) s.t. Lξ ω = 0 (i.e. the Lie derivative of ω by ξ vanishes).
Note that since ω is closed, this relation is equivalent ot d(ξ ω) = 0. An
important case occurs when ξ ω is exact: then there exists an (n − 1)-form
F s.t.

dF + ξ ω = 0. (6.12)
102 F. Hélein

Any (n − 1)-form F on M s.t. there exists a vector field ξ satisfying (6.12) is

called an observable (n − 1)-form. In the case where n = 1 then F is a function
and in fact any function on M is an observable 0-form because the symplectic
form is non degenerate. However if n ≥ 2 then an arbitrary (n − 1)-form on M
is not observable in general, but if it is so then the vector field ξ s.t. (6.12) holds
is unique: we shall denote it by ξF . Observable (n − 1)-forms can be integrated
over hypersurfaces in an n-curve to produce observable functionals. For that
purpose, given some Hamiltonian function H on M we define a slice  to be
codimension one submanifold of M s.t. for any Hamiltonian n-curve  the
intersection of  with  is transverse. We also assume that  is co-oriented,
which means that ∀m ∈  the 1-dimensional quotient space Tm M/Tm  is
oriented. Then we can endow  ∩  with an orientation and define

F : F −→ R
 
 −→ F
∩

Then one can recover two important notions in the semi-classical theory of
fields. First one can define a bracket between observable (n − 1)-forms F and
G by the formula

{F, G} := ξF ∧ ξG ω = ξF dG = −ξG dF.

Obviously {F, G} is also an (n − 1)-form. Moreover one can prove that it is

also observable and that ξ{F,G} = [ξF , ξG ] [23, 20]. Then the set of observable
(n − 1)-forms equipped with this ‘Poisson bracket’ becomes almost a Lie
algebra (it satisfies the antisymmetry relation {F, G} + {G, F } = 0, but not the
Jacobi identity; we have instead {{G, H }, F } + {{H, F }, G} + {{F, G}, H } =
d(ξF ∧ ξG ∧ ξH ω), which, in the case where n = 2 can be understood as a
Lie 2-algebra structure [1]). However we can define the bracket
   
F, G := {F, G} (6.13)
  

which coincides with the Poisson bracket on functionals on fields used by

physicists. We will also meet an interpretation of this bracket in the next
Section.
A second important notion is the relation between observable forms and
the dynamics. Indeed if  is a Hamiltonian n-curve and if F is an observable
(n − 1)-form then one can use the dynamical equation (6.7) with the vector
field ξF . It gives us, ∀m ∈ , ∀X ∈ "n Tm ,

dF (X) = −ω(ξF , X) = −dH(ξF )β(X). (6.14)

 Multisymplectic formalism and the covariant phase space 103

Hence we see that if dH(ξF ) vanishes, then dF | vanishes. This implies by

using Stokes theorem that the restriction of the functional  F to the set E
of Hamiltonian n-curves does not depend on  but on its homology class.
For that reason we say that an observable (n − 1)-form F is dynamical if
dH(ξF ) = 0.

In a premultisymplectic manifold (M, ω, β)

The definition of an observable
 (n − 1)-form F , of the bracket and of the
observable functionals  F can be adapted mutatis mutandis to the case of a
premultisymplectic manifold (M, ω, β). The difference is that in such a space
the dynamical condition dH(ξ ) = 0 is empty (think that M is the level set of
some Hamiltonian function H on a multisymplectic manifold, then the fact that
ξ is tangent to this level set forces it to be in the kernel of dH). Hence any
observable (n − 1)-form is a dynamical one.
Moreover if ω is exact, i.e. ω = dθ , we know that Hamiltonian n-curve are
critical points of the action (6.11). We can thus see that dynamical observ-
able (n − 1)-forms correspond to symmetries of the variational problems
and the conservation law dF | = 0 for a Hamiltonian n-curve  is noth-
ing but Noether’s first theorem [30, 28]. Indeed for any observable (n − 1)-
form F ,
LξF θ = d(ξF θ ) + ξF dθ = d(ξF θ ) + ξF ω = d(ξF θ − F ).

Hence LξF θ is exact, so that ξF is a symmetry of the action  θ up to a
divergence term. The conserved current is just F | .

6.1.7 Hamilton–Jacobi equations

The Hamilton–Jacobi equation for a Hamiltonian function H on a multisym-
plectic manifold of the form "n T ∗ Z (or on a submanifold of it) is the follow-
ing equation on an (n − 1)-form S on Z (i.e. a section of the vector bundle
"n−1 T ∗ Z −→ Z):
H(z, dSz ) = 0. (6.15)
Alternatively the unknown may be chosen to be λ := dS: we then require that λ
is a closed n-form on Z (or a section λ of "n T ∗ Z −→ Z s.t. λ∗ ω = 0) which
is a solution of λ∗ H = 0.
Then if for instance Z = X × Y, and if we denote by π the projection
 from
∗
X × Y to X , λ := π dS provides us with a null Lagrangian functional X λ on
X (i.e. a Lagrangian density which satisfies the Euler–Lagrange equation for
any map). In contrast with non relativistic quantum mechanics, the usefulness
104 F. Hélein

of this equation in the quantization of fields is not clear for the moment. One of
the interests of the Hamilton–Jacobi equation is that it allows one in principle
to prove under some circumstances that some solutions of the Euler–Lagrange
system of equations are global minimizers, by following a classical strategy
designed by K. Weierstrass and D. Hilbert (see [47, 6, 36]). This strategy is the
exact analogue in the general theory of calculus of variations of the theory of
calibrations used in minimal surfaces.
Note that one could impose extra conditions such as requiring that λ =
ds 1 ∧ · · · ∧ ds n , where s 1 , . . . , s n are functions on Z plus the fact that the
graph of λ is foliated by solutions to the Hamilton equations (this provides then
a generalization of the picture built by Hamilton in order to reconcile the Fermat
principle with the Huygens principle): this was achieved by Carathéodory [2]
in his theory (see §6.1.8).

6.1.8 Some historical remarks

The generalization of the Hamilton equations to variational problems with sev-
eral variables developed first along two directions. One of these is the question
of deciding whether a given solution to a variational problem is a minimum
of the action functional. This question was answered locally for 1-dimensional
variational problems by C.G.J. Jacobi (by following a remark of Legendre) in
1837 [22] by founding a method to check that the second variation is nonnega-
tive which is based on solutions to the so-called Jacobi equation. Note that this
method was extended to several variables by A. Clebsch [4] in 1859. Later on
a global, nonlinear version of these ideas was developed by K. Weierstrass and
D. Hilbert to prove the minimality of some solutions. This theory is connected
with another famous work of Jacobi of the same year (1837), who obtained the
Hamilton–Jacobi equation [21] by generalizing the work of Hamilton relating
the Fermat principle to the Huygens principle. In 1890 V. Volterra wrote two
papers [45, 46] where, to my knowledge for the first time3 , two different gen-
eralizations of the Hamilton system of equations to variational problems with
several variables were proposed. In [46] Volterra extended the Weierstrass–
Hilbert theory to variational problems with several variables. This theory was
further developed by G. Prange in 1915 [32] and by C. Carathéodory in his book
in 1929 [2] and is called today the Carathéodory theory. In 1934 H. Weyl [47],
inspired by Carathéodory’s theory, proposed a variant of it which is based on
the same theory as the one proposed by Volterra in [45] and that we described

3 This was followed by a work by L. Koenigsberger [27] in 1901, quoted by T. De Donder in [8],
which unfortunately I have difficulties to understand.
 Multisymplectic formalism and the covariant phase space 105

in §6.1.1. Today this theory is called the De Donder–Weyl one by many

authors4 . Its geometrical framework is the affine multisymplectic manifold
"n2 T ∗ Z.
A second direction was the notion of invariant integrals due to H. Poincaré
[33] and further developed by E. Cartan [3] in 1922: here one emphasizes the
relationship of Hamilton equations with the search for invariants which may
be functions or differential forms. This point of view is strongly related to
the covariant phase space theory (see §6.2.1 below). This theory was devel-
oped in full generality by T. De Donder [8] in 1935 and his main contribu-
tion was to deduce the extension of the affine (‘De Donder–Weyl’) theory to
Lagrangian densities depending on an arbitrary number of derivatives, i.e. to
the theory expounded in §6.1.2. Hence although Weyl’s and De Donder’s
contributions are almost simultaneous they are independent in their inspi-
ration: Weyl’s starting point was the so-called Carathéodory theory, moti-
vated by the search for generalizations to several variables of the Hamilton–
Jacobi equation, whereas De Donder’s starting point was the theory of integral
invariants.
The fact that a continuum of different theories may exist for a given varia-
tional problem was first understood by T. Lepage [29] in 1936 and completely
described by P. Dedecker in 1953 [6]. Today we can picture these various
theories as submanifolds of the universal multisymplectic manifold "n T ∗ Z
introduced by J. Kijowski [24] in 1974.
Recently the so-called De Donder–Weyl theory (but that we should call the
‘first Volterra theory’) has beed studied by many authors starting with the impor-
tant work by the Polish school around 1970, i.e. by W. Tulczjew, J. Kijowski,
W. Szczyrba and later on in many papers which are referred to in e.g. [17, 11].
However the Lepage–Dedecker theory has received much less attention (to my
knowledge it was only considered by J. Kijowski [24], F. Hélein, J. Kouneiher
[19, 20, 17] and M. Forger, S. V. Romero [10]), probably because of its com-
plexity. The latter theory leads however to interesting phenomena, particularly
for gauge theories [19, 17], since first class Dirac constraints simply disappear
there.
The modern formulation using the multisymplectic (n + 1)-form as the key
of the structure of the theory seems to start with the papers of J. Kijowski
[23], H. Goldschmidt and P. Sternberg [16] in 1973 and the introduction of
observable (n − 1)-forms apparentely goes back to the work of K. Gawédski
[15] in 1972.

4 Including, in previous papers, the author of this note, who until recently was unaware of the
work of Volterra.
106 F. Hélein

6.1.9 An example
Let X be the n-dimensional Minkowski space-time with coordinates

x = (x 0 , x 1 , . . . , x n−1 )

and consider the linear Klein–Gordon equation on X :

∂ 2ϕ
ϕ + m2 ϕ := − ϕ + m2 ϕ = 0, (6.16)
∂t 2
 ∂2
where t = x 0 and  := n−1 i=1 (∂x i )2 . We use the notations x :=
(x 1 , . . . , x n−1 ) ∈ Rn−1 and x = (x 0 , x) ∈ Rn and we define the Euclidean
scalar product x · y := x 1 y 1 + · · · + x n−1 y n−1 on Rn−1 and the Minkowski
product

x · y = ημν x μ y ν = x 0 y 0 − x 1 y 1 − · · · − x n−1 y n−1 = x 0 y 0 − x · y,

on X . The multisymplectic formulation of (6.16) takes place in M :=

"n T ∗ (X × R), equipped with the multisymplectic form

ω := de ∧ β + dpμ ∧ dϕ ∧ βμ .

Note that ω = dθλ , where

θλ := eβ + λpμ dϕ ∧ βμ − (1 − λ)ϕdpμ ∧ βμ ,

where λ ∈ R is a parameter to be fixed later. The Hamiltonian function on M

corresponding to solutions of (6.16) is
1 1
H(x, ϕ, e, p) := e + ημν pμ pν + m2 ϕ 2 .
2 2
To a solution ϕ of (6.16)we associate a Hamiltonian n-curve

 = {(x, ϕ(x), e(x), p(x)) | x ∈ X }

which satisfies
⎧
⎪
⎪ μν ∂ϕ
⎨ p μ (x) = η (x)
∂x ν
(6.17)
⎪
⎪ 1 ∂ϕ ∂ϕ 1
⎩ e(x) = − ημν (x) ν (x) − m2 ϕ(x)2 .
2 ∂x μ ∂x 2
We define E to be the set of Hamiltonian n-curves s.t. for all time t, x −→
ϕ(t, x) is rapidly decreasing at infinity.
We denote by Pn−1 H M the set of dynamical observable (n − 1)-forms
F and spH M := {ξ | Lξ ω = 0, dH(ξ ) = 0}. Note that (n − 1)-forms F in
H M are found by looking at vector fields ξ in spH M and by solving
Pn−1
 Multisymplectic formalism and the covariant phase space 107

ξ ω + dF = 0. They are of the form

F =ζ θ + F# ,
where ζ is a vector field on the Minkowski space-time X which is a generator
of the action of the Poincaré group and
∂#
F# := p μ #(x) − ϕημν (x) βμ ,
∂x ν
where # is a solution of (6.16). Note that moreover
∂ ∂# ∂ ∂# ∂
ξ# := ξF# = #(x) + ημν ν (x) μ − m2 ϕ#(x) + pμ μ (x) .
∂ϕ ∂x ∂p ∂x ∂e
In the following we shall denote by
∂
Pμ(λ) := θλ
∂x μ
and we observe that since L ∂x∂μ θλ = 0, we have dPμ(λ) + ∂x∂ μ ω = 0. Hence
ξPμ(λ) = ∂x∂ μ .
The brackets of two dynamical observable forms F, G ∈ Pn−1 H M are given
as follows: for any pair #, $ of solutions of (6.16),
∂# ∂$
{F# , F$ } = ημν (x)$(x) − #(x) ν (x) βμ . (6.18)
∂x ν ∂x
 
We observe that d{F# , F$ } = 0. Hence Pn−1 H M,  {·, ·} can be
 understood
as a kind of central extension of the Lie algebra spH M, [·, ·] and the Lie
sub algebra spanned by forms F# as an infinite dimensional analogue of the
Heisenberg algebra with central charges given by (6.18). Lastly
  ∂ ∂
Pμ(λ) , F# = L ∂x∂μ F# − d F# = F ∂x∂#μ − d F# (6.19)
∂x μ ∂x μ
 
and Pμ(λ) , Pν(λ) = 0.
 For the purpose of quantization we look at functionals of the form F =
 F# which are simultaneously eigenvectors of the linear operators
 
F −→ Pμ , F ,
(λ)


for μ = 0, . . . , n − 1. We find by using relation (6.19) that the eigenvector

λ = c#. This implies (by using the eigenvalue equation
∂#
equation reduces to ∂x
for μ = 0, . . . , n − 1) that #(x) = αeik·x . But because # should also be a
solution of (6.16) we must then have
ημν k μ k ν = (k 0 )2 − |k|2 = m2 . (6.20)
108 F. Hélein


(We remark that the maps {  Pλ(λ) , ·} play the role of the generators of a Cartan
subalgebra.) Let us denote by C the mass shell, i.e. the set of all k = (k 0 , k) ∈
R4 which are solutions of (6.20). This set actually splits into two connected
components according to the sign of k 0 : we let C + := {k ∈ C | k 0 > 0}. For any
k ∈ C + we define
i
αk := Fieik·x /√2π 3 = √ 3
eik·x (pμ − iϕk μ ) βμ
2π
−i −ik·x μ
αk∗ := F−ie−ik·x /√2π 3 = √ 3e (p + iϕk μ ) βμ .
2π
The vector fields associated to these observable forms are:
/ 0
eik·x ∂ μ ∂
  ∂
ξk := ξαk = √ 3 i −k + ημν p k − im ϕ
μ ν 2
,
2π ∂ϕ ∂pμ ∂e
/ 0
e−ik·x ∂ ∂   ∂
ξk∗ := ξαk∗ = √ 3 −i − k μ μ + ημν pμ k ν + im2 ϕ .
2π ∂ϕ ∂p ∂e

We then define the observable functionals

 
∗
ak := αk and ak := αk∗ .
 

As the notations suggest these functionals are the classical analogues of respec-
tively the annihilation and the creation operators. The advantage however is
that our functionals ak and ak∗ are independent of the coordinate system. We
can choose  to be the hyperplane x 0 = t = 0 and, for any function f , denote
by f |0 the restriction of f to . Then, for any  ∈ E we have
 1
i ∂ϕ ∂ϕ 20 (k),
ak () = √ 3
|0 (x) − ik 0 ϕ|0 (x) e−ik·x d x = i |0 (k) + k 0 ϕ|
2π R3 ∂t ∂t

where, for all function ψ on R3 ,


1
3(k) := √
ψ ψ(x)e−ik·x d x.
3
2π R3

Similarly we have:

−i ∂ϕ
ak∗ () = √ 3 |0 (x) + ik 0 ϕ|0 (x) eik·x d x
2π R ∂t
3

1
∂ϕ 20 (−k).
= −i |0 (−k) + k 0 ϕ|
∂t
 Multisymplectic formalism and the covariant phase space 109

Hence we deduce that

  1 1  
20 (k) = 1 ak () + a ∗ ()
ϕ| and
∂ϕ
|0 (k) = ak () − ak∗ () ,
2k 0 k ∂t 2i
where k = (k 0 , k) := (k 0 , −k). Thus denoting dμ(k) = 2k10 dk 1 dk 2 dk 3 =
1
2k 0
d k, we have

1 1  
ϕ(0, x) = √ 3 0
d k eik·x ak () + ak∗ ()
2π R3 2k

1  
= √ 3 dμ(k) ak ()e−ik·x + ak∗ ()eik·x
2π C +
and

−i 1  
∂ϕ
∂t
(0, x) = √ 3 d k eik·x ak () − ak∗ ()
2π R3 2

−i  
= √ 3 dμ(k)k 0 ak ()e−ik·x − ak∗ ()eik·x .
2π C +
Recall that these integrals can be interpreted as integrals over C + through
the parametrization R3  k −→ (k 0 , k) ∈ C + and that dμ is a measure on C +
invariant by the action of the Lorentz group. Note also
√ that in order to agree
with some textbooks one should add an extra factor k 0 inside the integrals.
By using the relation (6.18) we obtain, ∀k,  ∈ C + ,
−iei(k+)·x μ
{αk , α } = (k − μ )βμ ,
(2π )3
ie−i(k+)·x μ
{αk∗ , α∗ } = (k − μ )βμ ,
(2π )3
iei(k−)·x μ
{αk , α∗ } = (k + μ )βμ .
(2π )3
These brackets cannot be integrated over the slice  := {x 0 = 0} in the measure
theoretical sense5 , but one can make sense of their integrals as distributions
over the variables k ±  ∈ R3 :
{ak , a } = {ak∗ , a∗ } = 0, ∀k,  ∈ C + ,
and
{ak , a∗ } = i2k 0 δ( − k).
 
5 This contrasts with the integrals ∩ αk and ∩ αk∗ which exist if the restrictions to  of ϕ
and of its time derivative are Lebesgue integrable.
110 F. Hélein

A way to regularize these operators and their brackets is, by using functions
f, g ∈ L2 (C + ), to define
 
af := dμ(k)f (k)ak , and ag∗ := dμ(k)g(k)ak∗ .
C+ C+

Then

{af , ag∗ } = i dμ(k)f (k)g(k).
C+

6.2 The covariant phase space

6.2.1 A short historical review
The simplest version of the covariant phase space is the set E of solutions of a
Hamiltonian time evolution problem. In this case the Cauchy problem consists
in choosing some point m0 in the ordinary phase space (classically positions and
momenta) and some time t0 and in looking for solutions of the Hamiltonian
vector flow which coincide with m0 at time t0 . This problem has a unique
solution in all regular cases and this means that E is in one to one correspondence
with the set of initial data {m0 }. In other words to each time t0 there corresponds
a natural ‘Cauchy coordinates system’ on E, which is just the set of initial
conditions {m0 }. The key property is that the Hamiltonian flow preserves the
symplectic structure: this means all the symplectic structures induced by these
‘Cauchy coordinates systems’ on E coincides. Hence this defines a canonical
symplectic structure on E. The substitution of the ordinary phase space by the
space of solutions is a classical analogue of the transition from the Schrödinger
picture to the Heisenberg picture in quantum Mechanics: in the Schrödinger
picture the dynamics of a particle is described by the evolution of some time
dependent ‘state’ which is represented by a complex line in some complex
Hilbert space (the quantum analogue of a point in the ordinary phase space),
whereas in the Heisenberg picture the state (still a complex line in a complex
Hilbert space) does not evolve with time so that it may be interpreted as a
quantum analogue of a solution of the dynamical equations, i.e. of a point in
E (actually more precisely on a Lagrangian submanifold in the phase space,
according to A. Weinstein).
In Mechanics this concept is relatively old: the idea of contemplating the
space of solutions of a mechanical problem itself probably has roots in the
method of the variation de la constante of J.L. Lagrange and the notion of
‘Lagrange bracket’ is very close to the symplectic structure on the phase space.
The observation that this space carries an intrinsic symplectic structure was
 Multisymplectic formalism and the covariant phase space 111

clearly formulated by H. Poincaré [33] in his theory of invariant integrals

(invariants intégraux) and later further developed by E. Cartan [3] and fully
recognized by J.M. Souriau [39]. T. De Donder [7] extended the notion of
integral invariant to variational problems with several variables, being hence
very to from the notion of covariant phase space in this context, although it
is not clear that he realized it. Actually it seems difficult to decide when the
concept of covariant phase space in field theory emerged. My own guess is that
such an idea could also have been inspired by quantum field theory, since it
may be thought as the classical analogue of the Heisenberg picture in quantum
fields theory. First known works in this direction are the R.E. Peierls bracket
in 1952 [31], followed by the paper of I. Segal [37] in 1960. Peierls defined a
bracket on the set of solutions to a relativistic hyperbolic wave equation which
can be understood now as the restriction of the Poisson bracket associated to
the covariant phase structure on a certain class of functionals on the phase
space. Segal proved that the set of solutions of a non linear field relativistic
wave equation precisely carries a symplectic structure and proposed to use this
symplectic structure to quantize fields (and his paper is among the ones at the
origin of the geometric quantization method). This idea was later developed
in a more and more general framework by P. L. Garcı́a [12] in 1968, Garcı́a
and A. Pérez-Rendón [13] in 1971, H. Goldschmidt, S. Sternberg [16] in 1973.
In my opinion, the more accomplished presentation is the one by J. Kijowski
and W. Szczyrba [25] in 1976, which gives the first elementary but general
presentation of this structure, by using the multisymplectic formalism.
A more recent appearance of this idea can be found in the papers by C.
Crnkovic and E. Witten [5] and by G. Zuckerman [48], where the authors were
apparently unaware of the previous references and have rediscovered this prin-
ciple, being guided by the concept of the variational bicomplex of F. Takens
[40] and the work of A.M. Vinogradov [41]. This was followed by several
developments in the physical (e.g. [9]) and the mathematical literature, where
this principle is often referred to as the Witten covariant phase space. A general
presentation in the framework of the secondary calculus of Vinogradov was
done by E. Reyes [34] and L. Vitagliano in [42] and in relation to multisym-
plectic geometry (as in the present paper) by M. Forger and S.V. Romero in
[10].

6.2.2 The basic principle

We expound here briefly the principle of the covariant phase space using the
multsymplectic formalism. Our presentation will be heuristic and we refer to
[25, 18] for details. We assume that we are given a premultisymplectic manifold
112 F. Hélein

(M, ω, β) (see §6.1.4) and, as in §6.1.5, that ω is exact, i.e. ω = dθ , for some
n-form θ . We note E the set of Hamiltonian n-curves in (M, ω, β), i.e. the set
of oriented n-dimensional submanifolds  ⊂ M which satisfy (6.10). Given
some  ∈ E, the tangent space6 to E at  represents the set of infinitesimal
deformations δ of  which preserves the equation (6.10). Such a deformation
δ can be represented by a vector field ξ tangent to M defined along , i.e. a
section over  of j∗ T M, which is the pull-back image of the tangent bundle
T M by the embedding map j :  −→ M. Given δ, the vector field ξ is of
course not unique, since for any tangent vector field ζ on  (i.e. a section of
the subbundle T  ⊂ j∗ T M), ξ + ζ represents also δ. If so we write:
 
δ = ξ= ξ + ζ.
 

Moreover the condition on δ of being tangent to E forces ξ to be a solution

of the Jacobi equation:
 
∀v ∈ C ∞ (M, Tm M), v Lξ ω | = 0. (6.21)
Note that, although ξ is not a vector field defined on M (neither on a neigh-
bourhood of  in M) but only on , one can make sense of Lξ ω| because 
is a solution of (6.10).
Then for any slice  (see §6.1.6), any  ∈ E and δ ∈ T E, we define


 (δ) := ξ θ,
∩

where ξ is a section of j∗ T M
over  s.t. δ = ξ and ξ θ is the interior
product of θ by ξ . This hence defines a 1-form  on E

The dependence of  on 
This is the first natural question. For that purpose let us consider a smooth
1-parameter family of slices (t )t and compute the derivative:
 
d t d
 (δ) = ξ θ = L ∂t∂ (ξ θ )
dt dt t ∩ t ∩

∂ ∂
= d(ξ θ ) + d ξ ∧ θ .
t ∩ ∂t ∂t
But d(ξ θ ) = Lξ θ − ξ dθ and thus
∂ ∂   ∂
d(ξ θ) = Lξ θ − ξ ω.
∂t ∂t ∂t
6 Note that since E may not be a manifold in general, the usual definition of a tangent space
should be replaced by a suitable notion, see [25, 18].
 Multisymplectic formalism and the covariant phase space 113

However we can 4 assume

5 w.l.g. (see [18]) that the vector fields ∂t∂ and ξ admit
extensions s.t. ξ, ∂t = 0. Then the preceding relation gives us
∂

∂ ∂ ∂
d(ξ θ ) = Lξ θ −ξ ∧ ω.
∂t ∂t ∂t
Hence
 
d t ∂ ∂
 (δ) = Lξ θ − ξ∧ ω
dt t ∩ ∂t t ∩ ∂t

∂
+ d ξ∧ θ . (6.22)
t ∩ ∂t

First let us consider a smooth curve s −→ s ∈ E s.t. 0 =  and ds

dt
= δ.
Then the first term in the r.h.s. of (6.22) is equal to
  
∂ d ∂  dt
Lξ θ = θ  = δSdt (δ),
∂t ds 
t ∩ t ∩s ∂t s=0

where we have posed:


dt ∂
S dt () := θ.
t ∩ ∂t
Second we can assume w.l.g. (see [18]) that we can choose ∂t∂ in such a
way that it is tangent to . Let (X2 , . . . , Xn ) be a system of tangent vectors
on  s.t. ∀t, ∀m ∈ t ∩ , (X2 (m), . . . , Xn (m)) is a basis of Tm (t ∩ ) and
( ∂t∂ (m), X2 (m), . . . , Xn (m)) is a basis of Tm . Then if ψ is a n-volume form on
 s.t. ψ( ∂t∂ , X2 , . . . , Xn ) = 1, the second term in the r.h.s. of (6.22) reads
 
∂ ∂
− ξ∧ ω=− ω ξ, , X2 , . . . , Xn ψ
t ∩ ∂t t ∩ ∂t
and vanishes because of the Hamilton equations (6.10). Lastly we assume that
the restriction of ξ to t ∩  has compact support or is rapidly decreasing: this
occurs for instance if the Hamilton system encodes hyperbolic wave equations,
if  is a level hypersurface of some time coordinate and if we impose that the
Hamiltonian n-curves in E have a prescribed behaviour at infinity in space for
all time. Then the last term in the r.h.s. of (6.22) vanishes. Then Relation (6.22)
can be rewritten
d t
dt

 (δ) = δSdt (δ), ∀δ ∈ T E

dt
or
d   dt
t
= δS dt . (6.23)
dt
114 F. Hélein

We can also define the functional


S12 () := θ,
∩{t1 ≤t≤t2 }

which represents the ‘action’ between the slices 1 := {t = t1 } an 2 := {t =

t dt
t2 }. Then S12 () = t12 S dt ()dt and thus we deduce by integrating (6.23)
over [t1 , t2 ] that
2
− 1
= δS12 . (6.24)

The symplectic form

In view of the preceding we are led to the conclusion that, although the 1-form

depends on , its differential δ  does not depend on  since (6.24) tells
us that 2 − 1 is an exact form. Of course one should be careful in using
the identity δ ◦ δ = 0 since E is not a smooth manifold (see [18] for a rigorous
proof that δ  does not depend on ). All this motivates the definition of the
following 2-form on E:

 := δ .

We will prove that  has the following expression: ∀δ1 , δ2  ∈ T E,


 (δ1 , δ2 ) = ξ1 ∧ ξ2 ω, (6.25)
∩
 
where ξ1 , ξ2 are sections over  of j∗ T M s.t. δ1  =  ξ1 and δ2  =  ξ2 .
To prove (6.25) we need to compute δ   (δ1 , δ2 ). For that purpose we first
assume that we can extend the two tangent vectors δ1  and δ2  to commuting
vector fields on E around  (actually we can assume that [ξ1 , ξ2 ] = 0). Then

δ 
 (δ1 , δ2 ) = δ1  · 
 (δ2 ) − δ2  ·  (δ1 )

 −  ([δ1 , δ2 ])


= δ1  · ξ2 θ − δ2  · ξ1 θ .
∩ ∩

Thus
 
δ 
 (δ1 , δ2 ) = Lξ1 (ξ2 θ) − Lξ2 (ξ1 θ) .
∩ ∩

We use then the following identity (see [18]): for any pair of vector fields X1
and X2 and for any p-form β,

LX1 (X2 β)−LX2 (X1 β) = X1 ∧ X2 dβ +[X1 , X2 ] β +d(X1 ∧ X2 β).

 Multisymplectic formalism and the covariant phase space 115

Setting X1 = ξ1 , X2 = ξ2 and β = θ , we obtain using [ξ1 , ξ2 ] = 0 and dθ = ω

that Lξ1 (ξ2 θ ) − Lξ2 (ξ1 θ ) = ξ1 ∧ ξ2 ω − d(ξ1 ∧ ξ2 θ ). Thus

δ  (δ1 , δ2 ) =

ξ1 ∧ ξ2 ω − d(ξ1 ∧ ξ2 θ ). (6.26)
∩

Hence if we assume that the restriction of ξ1 and ξ2 to t ∩  has compact

support or is rapidly decreasing (as in the preceding paragraph) we obtain
(6.25).
Hence we conclude that, under some hypotheses, one can endow the set
E of solutions to the Hamilton equations with a symplectic form  given by
(6.25). This form
 does depend
 not on  under
 the condition that the boundary
terms t ∩ d ξ ∧ ∂t∂ θ in (6.22) and − ∩ d(ξ1 ∧ ξ2 θ ) in (6.26) vanish.
This means that, on each slice , the Jacobi vector fields ξ, ξ1 , ξ2 decrease
sufficiently rapidly at infinity. Such a condition is true if, for instance, the
manifold X is a Lorentzian manifold, the slice  is (a lift of) a spacelike
hypersurface of X and we impose in the definition of E that all Hamiltonian
n-curves  in E are asymptotic to a given ‘ground state’ Hamiltonian n-curve
0 at infinity on each slice .
With such a symplectic structure  on E we can define a Poisson bracket on
real-valued functionals on E, which is nothing but (6.13).

6.2.3 A geometric view of the proof

We can give an alternative proof of Relation (6.24) with a more geometric
flavor. We will be even more heuristic, however the validity of our argument
is strongly based on the fact that the Lagrangian action can be represented by
(6.11). For that purpose imagine that our problem models a hyperbolic time
evolution problem and that there are well-defined notions of time and space
coordinates on M (as is the case for any wave equation on a curved space-time).
Consider a Hamiltonian n-curve  and let   be another Hamiltonian n-
curve, which we suppose to be close to . More precisely we assume that
  =  + εδ + o(ε), where ε > 0 is a small parameter : by thiscondition we
mean that there exists a vector field ξ ∈ C ∞ (M, T M) s.t. δ =  ξ and   is
the image of  by the flow map eεξ . We also assume that, for all ‘time’,   is
asymptotic to  at infinity in space. Let 1 and 2 be two slices, assume that
these slices are space-like hypersurfaces and, in order to fix ideas, we suppose
that 2 is in the future of 1 . These slices cross transversally  and   and
we denote by σ1 (resp. σ2 ) the piece of 1 (resp. 2 ) which is enclosed by the
intersections with  and   (see the picture). We also denote by − the part of
 which is in the past of 1 , by + the part of  which is in the future of 2 and
116 F. Hélein

Γ+

∑1
σ2

Γ0 Γ0’

∑2
σ1

Γ–

1 2
Figure 6.1 A geometric comparison of (δ) with (δ)

by 0 the part of   which is between 1 and 2 (see again the picture). Lastly
we consider the (not necessarily Hamiltonian) n-curve ε , which is the union
of − , σ1 , 0 , σ2 and + . Of course ε is not smooth, but it can be approached
by a sequence of smooth n-curves, so that the following makes sense. We also
endow ε with the orientation which agrees with that of  on − ∪ + and
with that   on 0 .
Let us use the fact that  is a Hamiltonian n-curve, hence a critical point of
(6.11). This implies that
 
θ= θ + o(ε). (6.27)
ε 

However the l.h.s. of (6.27) can be decomposed as

     
θ= θ+ θ+ θ+ θ+ θ,
ε − σ1 0 σ2 +

whereas its r.h.s. is

   
θ + o(ε) = θ+ θ+ θ + o(ε),
 − 0 +
 Multisymplectic formalism and the covariant phase space 117

where 0 is the part of  between 1 and 2 . Hence (6.27) reduces to

   ! 
θ+ θ− θ + θ = o(ε). (6.28)
σ1 0 0 σ2

 
We now recognize that, on the one hand, σ1 θ = ε 1 (δ) + o(ε), σ2 θ =
−ε 2 (δ) + o(ε) (the sign
 being due to the orientation of σ2 ). On the other
hand 0 θ = S12 (0 ) and  θ = S12 (0 ) = S12 (0 + εδ0 ) + o(ε). Hence
0
(6.28) gives us

ε 1
(δ) + ε δS12 (δ) − ε 2
(δ) = o(ε).


Thus by dividing by ε and letting ε tend to 0, we recover (6.11).

6.3 Geometric quantization

We address here the question of building a geometric quantization scheme, or
at least a prequantization scheme for fields based on the covariant phase space
structure. This was more or less the programme envisioned by G. Segal in 1960
[37]. We present here an attempt at that by using the multisymplectic theory on
a very elementary example, which is the one presented in §6.1.9.

Canonical vector fields on the set of solutions E

H M a tangent vector field F on E wich is
We can associate to each F ∈ Pn−1
given by

∀ ∈ E, F () := ξF .


In the case of the Klein–Gordon equation (6.16) it is interesting to represent

solutions in E by local coordinates. The most convenient way is based on the
Fourier transform: any Hamiltonian n-curve  is characterized by a solution ϕ
to (6.16) and by writing

1  
ϕ(x) = √ 3 dμ(k) uk e−ik·x + u∗k eik·x , (6.29)
2π C +
we get formally a map
+ +
E −→ CC × CC
ϕ −→ (uk , u∗k )k∈C + .
118 F. Hélein

Note that the image of E is characterized by the reality condition uk = u∗k ,

∀k ∈ C + . We can obviously extend this map to the complexification E C of E
and then this map is an isomorphism.

The creation and annihilation canonical transformations

Now given some function f ∈ L2 (C + ) consider
 
i
αf := dμ(k)f (k)αk = √ 3 dμ(k)eik·x f (k)(pμ − iϕk μ )βμ .
C+ 2π C +
(Note that the observable functional af defined in §6.1.9 is obtained by inte-
gration of αf over a slice.) Then

i ∂ ∂ ∂
ξf := ξαf = √ 3 dμ(k)eik·x f (k) ik μ μ −(m2 ϕ +iημν pμ k ν ) +
2π C + ∂p ∂e ∂ϕ

is completely characterized by the fact that it preserves ω and dH and through

its action on ϕ:

  i
dϕ ξf = √ 3 dμ(k)eik·x f (k).
C +
2π
We can easily integrate ξf on M and its action on F:

i
U (s, ϕ) = ϕ + s √ 3 dμ(k)eik·x f (k).
2π C +
In terms of the coordinates (uk , u∗k )k∈C + it gives:

U (s, uk , u∗k ) = (uk , u∗k + isf (k)).

Hence we can symbolically denote


∂
f := αf = i f (k) .
C+ ∂u∗k

There is no integration measure used here, the sign stands uniquely for
summing linearly independent vectors: the meaning is that
 ! 
1 d k  −ik·x ∗ ik·x
 1
f √ 3 0
uk e + uk e =√ 3 dμ(k)if (k)eik·x .
2π C + 2k 2π C +
A completely analogous computation can be done for
 
−i
αg∗ := dμ(k)g(k)αk∗ = √ 3 dμ(k)e−ik·x g(k)(pμ + iϕk μ )βμ ,
C + +
2π C
 Multisymplectic formalism and the covariant phase space 119

where g ∈ L2 (C + ). Denoting ξg∗ := ξαg∗ we have


 ∗ −i
dϕ ξg = √ 3 dμ(k)e−ik·x g(k).
2π C +
Hence ∗g := αg∗ is given by

∂
∗g = −i g(k) .
C+ ∂uk

Spacetime translations
We now look at the canonical vector fields on E associated with spacetime
translations Pζ , where ζ is constant vector field on X . We recall that ξP (λ) = ζ .
ζ
We must understand the induced vector field ζ on F. Let U (s, ·) be the flow
mapping of the vector field ζ : U (s, x, ϕ, e, p) = (s, x + ζ, ϕ, e, p). Then the
image of

 = {(x, ϕ(x), e(x), p(x)) | x ∈ X } ⊂ E

by U (x, ·) is

s = {(x, ϕ(x − sζ ), es (x), ps (x)) | x ∈ X },

where the value of es (x) and ps (x) is completely determined by the constraint
that s ⊂ E and by the knowledge of ϕ(x − sζ ). This can be proved by a simple
change of variable. Similarly we determine the action of ζ on the coordinates
(uk , u∗k )k∈C + by computing its action on ϕ:
 !
   −ik·(x−sζ )  
d 1 
ζ ϕ (x) = √ 3 + dμ(k) uk e + u∗k eik·(x−sζ ) 
ds 2π C 
s=0

1  
= √ 3 dμ(k) ik · ζ uk e−ik·(x−sζ ) − ik · ζ u∗k eik·(x−sζ )
+
2π C
/  0
∂ ∗ ∂
=i k · ζ uk − uk ∗ ϕ (x).
C+ ∂uk ∂uk
Hence

∂ ∂
ζ = i k · ζ uk − u∗k ∗ .
C+ ∂uk ∂uk

Geometric prequantization
We recall very briefly the prequantization scheme due to B. Kostant and J.-M.
Souriau (generalizing previous constructions by B.O. Koopman, L. Van Hove
and I. Segal, see [26, 38]). We let (M, ω) be a simply connected symplectic
120 F. Hélein

manifold and we assume for simplicity that there exists a 1-form θ with ω = dθ .
We consider the trivial bundle L := M × C and denote by (M, L) the set of
square integrable sections of L. Using θ we can define a Hermitian connection
∇ acting on (M, L) by

i
∀ξ ∈ (M, T M), ∀ψ ∈ (M, L), ∇ξ ψ = ξ · ψ − θ (ξ )ψ.

3 acting on
Then to each function F ∈ C ∞ (M, R) we associate the operator F
(M, L)

3ψ = F ψ +  ∇ξF ψ = (F − θ (ξF )) ψ +  ξF · ψ,
F
i i
where dF + ξF ω = 0. This construction is called the prequantization of
(M, ω). For instance if (M, ω) = (R2n , dpi ∧ dq i ), then ω = dθ , with θ =
pi dq i and q3i = q i + i ∂p∂ i and p
3i = −i ∂q∂ i . Of course one needs further
restrictions in order to recover an irreducible representation of the Heisenberg
algebra (and hence the standard quantization): this will be the purpose of
introducing a polarization and a tensorization of the line bundle L with the
bundle of half volume forms transversal to the leaves of the polarization (see
[26, 38]).
We will propose an extension of this procedure to our setting, concerned
with the quantization of fields. We consider the trivial bundle L := E C × C
over E C , where E C is the complexification of the set of solutions to the Klein–
Gordon equation (6.16) as before. On the set (E C , L) of smooth sections of
L (we are here relatively vague about the meaning of "smooth") we define a
notion of covariant derivative along any vector field of the type F , where
F ∈ (E C , L) by

C i
∀ψ ∈ (E , L), ∇F ψ := F · ψ − ξF θ ψ,
 

where θ = θλ . Then we define the prequantization of F ∈ Pn−1

H M to be the
operator acting on (E C , L) by:
 
3ψ :=  
F F ψ+ ∇F ψ = F · ψ + F − ξF θ ψ.
 i i 

Prequantization of the creation and annihilation observables

We look here for the expressions of the prequantization of af and ag∗ given in
§6.1.9. We first note the fact that if ϕ is given in terms of (uk , u∗k )k∈C + by (6.29),
 Multisymplectic formalism and the covariant phase space 121

then
 uk + u∗k
1
√ 3 e−ik·x ϕ(0, x)d x = ,
2π Rn 2k 0

and
 uk − u∗k
1
√ 3 e−ik·x p0 (0, x)d x = ,
2π Rn 2i

where k = (k 0 , −k). We deduce the following

  
1 d k −ik·x  
αf = √ 3 dx 0
e f (k) ϕ(0, x)k 0 + ip0 (0, x)
∩ 2π Rn C + 2k
 uk + u∗k uk − u∗k
dk
= 0
f (k) +
C + 2k 2 2

dk
= 0
f (k)uk .
C + 2k

Similarly
 
dk
αg∗ = g(k)u∗k .
∩ C+ 2k 0

We moreover observe that


1 d k ik·x f (k) αf
ξf θ=√ 3
e (ϕk μ + ipμ ) βμ = ,
2π C+ 2k 0 2 2

αg∗
and similarly ξg∗ θ= 2
. Hence
   
d k f (k) d k g(k) ∗
ξf θ= uk and ξg∗ θ= u .
∩ C+ 2k 0 2 ∩ C+ 2k 0 2 k

Using the previous results we can now express, for ψ ∈ (E C , L),

i
∇f ψ := f · ψ − ξf θ ψ
 
 
∂ψ i d k f (k)
= i f (k) ∗ − uk ψ
C+ ∂uk  C + 2k 0 2
122 F. Hélein

and

i
∇∗g ψ := ∗g · ψ − ξ∗ θ ψ
  g
 
∂ψ i d k g(k) ∗
= −i g(k) − uk ψ.
C + ∂uk  C + 2k 0 2

For the prequantizations we obtain:

 
∂ψ d k f (k)
3
af ψ =  f (k) + uk ψ,
C+ ∂u∗k C+ 2k 0 2

and
 
∂ψ d k g(k) ∗
ag∗ ψ = −
3 g(k) + u ψ.
∂uk C+ 2k 0 2 k

We observe that we have formally [3 af  ] = [3

af ,3 ag∗ ] = 0 and
ag∗ ,3


dk
[3 ag∗ ] = 
af ,3 f (k)g(k).
C+ 2k 0

Prequantization of the stress-energy tensor

It relies on finding the prequantization
 of Pζ(λ) = ζ μ Pμ(λ) . In principle one
(λ)
should compute the functionals of  Pζ and  ξP (λ) θλ . But as observed in
ζ
(λ)
the previous
 section we have P ζ = ζ θλ = ξ P (λ) θλ because Lieζ θλ = 0.
ζ
Hence  Pζ − ξP (λ) θλ = 0 and so the prequantization of Pζ(λ) is just
(λ)
ζ


 ∂ ∂
P1
(λ) ψ =
ζ ζ · ψ =  k · ζ uk − u∗k ∗ ψ.
i C+ ∂uk ∂uk


Note that if we need to compute  Pζ(λ) , it is more suitable to set λ = 1, since
it gives then the standard expression for the stress-energy tensor. For instance,
if ζ = ∂x∂ 0

  3
!
(p 0 )2 (pi )2 ϕ2
− P0(1) = dx + + m2
∩ Rn−1 2 i=1
2 2

gives the total energy in the frame associated with the coordinates x μ .
 Multisymplectic formalism and the covariant phase space 123

Introducing a polarization
We choose to impose the extra condition ∇∗g ψ = 0, ∀g (covariantly antiholo-
morphic sections), which gives us:

1 dk
ψ(uk , u∗k ) = h(u∗k ) exp − uk u∗k = h(u∗k )|0.
2 C + 2k 0

The advantage of this choice is that all observable functionals (creation and
annihilation, energy and momentum) are at most linear in the variables (uk , u∗k ),
so we do not need to use the Blattner–Kostant–Sternberg correction for these
3|0 = 0, so that the energy of the vacuum van-
operators [26, 38]. As a result P
ishes without requiring normal ordering. However we did not take into account
the metaplectic correction, which requires a slight change of the connection:
were we to do so we would find that the vacuum has infinite energy (as in
the standard quantization scheme), which can be removed by a normal order-
ing procedure. The mysterious thing here (as was already observed) is that by
ignoring the metaplectic correction (which however is fundamental for many
reasons) we do not need the normal ordering correction.

References
[1] J. C. Baez, C. L. Rogers, Categorified Symplectic Geometry and the String Lie
2-algebra, Homology, Homotopy Appl. 12 (2010), no. 1, 221–236. arXiv:0901.
4721.
[2] C. Carathéodory, Variationsrechnung und partielle Differentialgleichungen erster
Ordnung, Teubner, Leipzig (reprinted by Chelsea, New York, 1982); Acta litt. ac
scient. univers. Hungaricae, Szeged, Sect. Math., 4 (1929), p. 193.
[3] E. Cartan, Leçons sur les invariants intégraux, Hermann, 1922.
[4] A. Clebsch, Ueber die zweite Variation vielfache Integralen, J. reine angew. Math.
56 (1859), 122–148.
[5] C. Crnkovic, E. Witten, Covariant description of canonical formalism in geo-
metrical theories, in Three hundred years of gravitation (S. W. Hawking and
W. Israel, eds.), Cambridge University Press, Cambridge, 1987, 676–684; E.
Witten, Interacting field theory of open supertrings, Nucl. Phys. B 276 (1986),
291–324.
[6] P. Dedecker, Calcul des variations, formes différentielles et champs géodésiques,
in Géométrie différentielle, Colloq. Intern. du CNRS LII, Strasbourg 1953, Publ.
du CNRS, Paris, 1953, p. 17–34; On the generalization of symplectic geometry
to multiple integrals in the calculus of variations, in Differential Geometrical
Methods in Mathematical Physics, eds. K. Bleuler and A. Reetz, Lect. Notes
Maths. vol. 570, Springer-Verlag, Berlin, 1977, p. 395–456.
[7] T. De Donder, Introduction à la théorie des invariants intégraux, Bull. Acad. Roy.
Belgique (1913), 1043–1073.
124 F. Hélein

[8] T. De Donder, Théorie invariantive du calcul des variations, Gauthiers-Villars,

Paris, 1930.
[9] B.P. Dolan, K.P. Haugh, A co-variant approach to Ashtekar’s canonical gravity,
Class. Quant. Gravity, Vol. 14, N. 2, (1997), 477–488 (12).
[10] M. Forger, S. V. Romero, Covariant Poisson bracket in geometric field theory,
Commun. Math. Phys. 256 (2005), 375–410.
[11] M. Forger, L. Gomes, Multisymplectic and polysymplectic structures on fiber
bundles, preprint arXiv:0708.1586.
[12] P. L. Garcı́a, Geometrı́a simplética en la teoria de campos, Collect. Math. 19, 1–2,
73, 1968.
[13] P. L. Garcı́a, A. Pérez-Rendón, Symplectic approach to the theory of quantized
fields, I, Commun. Math. Phys. 13 (1969), 24–44 and —, II, Arch. Rational Mech.
Anal. 43 (1971), 101–124.
[14] M.J. Gotay, J. Isenberg, J.E. Marsden (with the collaboraton of R. Montgomery,
J. Śnyatycki, P.B. Yasskin), Momentum maps and classical relativistic fields,
Part I/covariant field theory, preprint arXiv/physics/9801019
[15] K. Gawȩdski, On the generalization of the canonical formalism in the classical
field theory, Rep. Math. Phys. No 4, Vol. 3 (1972), 307–326.
[16] H. Goldschmidt, S. Sternberg, The Hamilton–Cartan formalism in the calculus of
variations, Ann. Inst. Fourier Grenoble 23, 1 (1973), 203–267.
[17] F. Hélein, Hamiltonian formalisms for multidimensional calculus of variations
and perturbation theory, in Noncompact problems at the intersection of geometry,
analysis, and topology, Contemp. Math., 350 (2004), 127–147.
[18] F. Hélein, The use of the covariant phase space on non nonlinear fields, in prepa-
ration.
[19] F. Hélein, J. Kouneiher, Covariant Hamiltonian formalism for the calculus of
variations with several variables: Lepage–Dedecker versus De Donder–Weyl, Adv.
Theor. Math. Phys. 8 (2004), 565–601.
[20] F. Hélein, J. Kouneiher, The notion of observable in the covariant Hamiltonian
formalism for the calculus of variations with several variables, Adv. Theor. Math.
Phys. 8 (2004), 735–777.
[21] C.G.J. Jacobi, Ueber die Reduction der Integration des partiellen Differentialgle-
ichungen erster Ornung zwischen irgend einer Zahl Variabeln auf die Integration
eines einzigen Systemes gewöhnlicher Differentialgleichungen, J. reine angew.
Math. 17 (1837), 68–82.
[22] C.G.J. Jacobi, Zur Theorie des Variations-Rechnung und des Differential-
Gleichungen, J. reine angew. Math. 17 (1837), 97–162.
[23] J. Kijowski, A finite dimensional canonical formalism in the classical field theory,
Commun. Math. Phys. 30 (1973), 99–128.
[24] J. Kijowski, Multiphase spaces and gauge in the calculus of variations, Bull. de
l’Acad. Polon. des Sci., Série sci. Math., Astr. et Phys. XXII (1974), 1219–1225.
[25] J. Kijowski, W. Szczyrba, A canonical structure for classical field theories, Com-
mun. Math Phys. 46 (1976), 183–206.
[26] A.A. Kirillov, Geometric quantization, in Dynamical systems IV, V.I. Arnol’d, S.P.
Novikov, eds., Springer-Verlag, 1990.
[27] L. Koenigsberger, Die Prinzipien der Mechanik für mehrere Variable, Sitzungs-
berichte Akad. Wiss. Berlin, Bd. XLVI, 14 nov. 1901, 1108; Die Prinzipien der
 Multisymplectic formalism and the covariant phase space 125

Mechanik für mehrere unavhängige Variable, J. Reine Angw. Math., Bd. 124
(1902), 202–277.
[28] Y. Kosmann-Schwarzbach, Les théorèmes de Noether – Invariance et lois de
conservation au XXe siècle, Les éditions de l’Ecole Polytechnique, 2004.
[29] T. Lepage, Sur les champs géodésiques du calcul des variations, Bull. Acad. Roy.
Belg., Cl. Sci. 27 (1936), 716–729, 1036–1046.
[30] E. Noether, Invariante Variationsprobleme, Nachrichten von der Königlichen
Gesellschaft des Wissenschaften su Göttingen, Mathematisch-physikalische
Kalsse, 1918, p. 235–257.
[31] R.E. Peierls, The commutation laws of relativistic field theory, Proc. Roy. Soc.
London, Ser. A, Vol. 214, No. 1117 (1952), 143–157.
[32] G. Prange, Die Hamilton–Jacobische Theorie für Doppelintegrale, Diss.
Göttingen, 1915.
[33] H. Poincaré, Les méthodes nouvelles de la mécanique céleste, t. III, Paris, Gauthier–
Villars, 1899.
[34] E.G. Reyes, On covariant phase space and the variational bicomplex, Int. J. Theor.
Phys., Vol. 43, No. 5 (2004), 1267–1286.
[35] C. Rovelli, A note on the foundation of relativistic mechanics – II: Covari-
ant Hamiltonian general relativity, in Topics in Mathematical Physics, Gen-
eral Relativity and Cosmology, H Garcia-Compean, B Mielnik, M Montesinos,
M Przanowski editors, pg 397, (World Scientific, Singapore 2006). arXiv:gr-
qc/0202079
[36] H. Rund, The Hamilton–Jacobi theory in the calculus of variations, its role in
mathematics and physics, Krieger Pub. 1973.
[37] I. Segal, Quantization of nonlinear systems, J. Math. Phys. vol. 1, N. 6 (1960),
468–488.
[38] J. Śnyatycki, Geometric quantization and quantum Mechanics, Appl. Math. Sci.
30, Springer-Verlag 1980.
[39] J.-M. Souriau, Structure des systèmes dynamiques, Dunod, Paris, 1970.
[40] F. Takens, A global formulation of the inverse problem of the calculus of variations,
J. Diff. Geom. 14 (1979), 543–562.
[41] A.M. Vinogradov, The C-spectral sequence, Lagrangian formalism, and conser-
vations laws, I and II, J. Math. Anal. Appl. 100 (1984), 1–40 and 41–129.
[42] L. Vitagliano, Secondary calculus and the covariant phase space, J. Geom. Phys.
59 (2009), no. 4, 426–447.
[43] L. Vitagliano, The Lagrangian–Hamiltonian Formalism for Higher Order Field
Theories, J. Geom. Phys. 60 (2010), no. 6-8, 857–873. arXiv:0905.4580.
[44] L. Vitagliano, Partial Differential Hamiltonian Systems, preprint arXiv:0903.4528
[45] V. Volterra, Sulle equazioni differenziali che provengono da questiono di calcolo
delle variazioni, Rend. Cont. Acad. Lincei, ser. IV, vol. VI, 1890, 42–54.
[46] V. Volterra, Sopra una estensione della teoria Jacobi–Hamilton del calcolo delle
varizioni, Rend. Cont. Acad. Lincei, ser. IV, vol. VI, 1890, 127–138.
[47] H. Weyl, Geodesic fields in the calculus of variation for multiple integrals, Ann.
Math. (3) 36 (1935), 607–629.
[48] G. Zuckerman, Action functional and global geometry, in Mathematical aspects
of string theory, S.T. Yau, eds., Advanced Series in Mathematical Physics, vol 1,
World Scientific, 1987, 259–284.
126 F. Hélein

Author’s address:

Institut de Mathématiques de Jussieu,

UMR CNRS 7586 Université Denis Diderot Paris 7,
175 rue du Chevaleret,
75013 Paris, France
[email protected]
 7
Nonnegative curvature on disk bundles
lorenz j. schwachh öfer

7.1 Introduction
The search for manifolds of nonnegative curvature1 is one of the classical
problems in Riemannian geometry. While general obstructions are scarce, there
are relatively few general classes of examples and construction methods. Hence,
it is unclear how large one should expect the class of closed manifolds admitting
a nonnegatively curved metric to be. For a survey of known examples, see
e.g. [12].
Apart from taking products, there are only two general methods to construct
new nonnegatively curved metrics out of given spaces. One is the use of Rie-
mannian submersions which do not decrease curvature by O’Neill’s formula.
The other is the glueing of two manifolds (which we call halves) along their
common boundary. Typically, the boundary of each half is assumed to be totally
geodesic or, slightly more restrictively, a collar metric. This in turn implies by
the Soul theorem ([2]) that each half is the total space of a disk bundle over a
totally geodesic closed submanifold. In addition, the glueing map of the two
boundaries must be an isometry.
While many examples can be constructed by such a glueing, its application
is still limited. On the one hand, there is not too much known on the question
of which disk bundles over a nonnegatively curved compact manifold admit
collar metrics of nonnegative curvature, and on the other hand, even if such
metrics exist, the metric on the boundary is not arbitrary. Thus, glueing together
two such disk bundles to a nonnegatively curved closed manifold is possible in
special situations only.
For instance, if the disk bundle is homogeneous, then there always exist
invariant nonnegatively curved collar metrics. However, the metric on the

1 Throughout this article, the term “curvature” refers to the sectional curvature.

127
128 L. J. Schwachhöfer

boundary of such a collar metric is restricted due to the existence of certain

parallel Killing fields by a result of Perelman ([7]).
In this article, we will give a survey of known examples and describe some
recent results which illustrate the difficulty in finding metrics on disk bundles
which are suitable for this glueing construction.

7.2 Normal homogeneous metrics and Cheeger deformations

Let G be a compact Lie group with Lie algebra g, and let g0 be a biinvariant
metric on G, i.e., a metric for which all left and right translations are isometries.
It is well known that each compact Lie group carries such a metric, and its
curvature is given by
1
secg0 (x, y) = |[x, y]|2g0 ≥ 0
4
for each orthonormal pair x, y ∈ Tg G = dLg g ∼ = g. Let H ⊂ G be a closed
subgroup, and consider the compact homogeneous space M := G/H . Then
there is a unique G-invariant metric on G/H which by abuse of notation we
also denote by g0 , for which the canonical projection G → G/H becomes
a Riemannian submersion. This metric on G/H is called a normal homoge-
neous metric. Evidently, a normal homogeneous metric always has nonnegative
curvature by O’Neill’s formula.
Let (M, g) be a Riemannian manifold, and suppose that G acts isometrically
on M. The action map
G × M −→ M
may be regarded as a principal G-bundle with the free action of G on G × M
given by
k % (g, p) := (gk −1 , k · p)
for g, k ∈ G and p ∈ M. Evidently, this action is isometric with respect to the
product of a biinvariant metric on G and the given metric on M.
Definition 7.2.1 Let (M, g) be a Riemannian manifold, and let G be a Lie
group with a biinvariant metric g0 which acts isometrically on M. Then for
λ > 0, the metric gλ on M for which the action map
(G × M, λg0 ⊕ g) −→ (M, gλ )
becomes a Riemannian submersion is called a Cheeger deformation of g.
Since Cheeger deformations are Riemannian submersions and hence curva-
ture non-decreasing by O’Neill’s formula, it follows that gλ has nonnegative
 Nonnegative curvature on disk bundles 129

curvature whenever g does. Moreover, gλ may have nonnegative curvature even

if g has some negative curvature.
Proposition 7.2.2 Let g be a G-invariant metric on M and let gλ denote the
Cheeger deformation of g for λ > 0. Then
(i) limλ→∞ gλ = g.
(ii) (gλ )μ = g λμ . That is, the Cheeger deformation of a Cheeger deformation
λ+μ
is again a Cheeger deformation.
(iii) If gλ0 has nonnegative curvature for some λ0 > 0, then so does gλ for any
λ < λ0 .
(iv) If M = G/H is a homogeneous space with a G-invariant metric g, then
limλ→0 λ1 gλ = g0 , where g0 is the normal homogeneous metric on G/H
induced by the biinvariant metric g0 on G.
Proof Let p ∈ M, and let N := G · p ⊂ M be the G-orbit through p. We
decompose the tangent space g-orthogonally as
Tp M = Tp N ⊕ Sp .
We write N = G/H with H := Stab(p) ⊂ G, and choose the orthogonal
decomposition of the Lie algebra g = h ⊕ m, so that m ∼ = Tp N with the identi-
fication x → (x ∗ )p , where x ∗ denotes the action field of x ∈ g. Let ϕ : m → m
be the self-adjoint map for which
g(x ∗ , y ∗ )p = g0 (ϕ(x), y),
where g0 is the given biinvariant inner product on g. The tangent space of the
fiber of the submersion G × M → M at (e, p) is given by {(−x, (x ∗ )p ) | x ∈
g}, hence its orthogonal complement with respect to (λg0 ) ⊕ g is
H(e,p) = {(ϕ(x), λ(x ∗ )p ) | x ∈ m} ⊕ Sp .
Therefore, the horizontal lift of (x ∗ )p is (ϕ(x̂), λ(x̂ ∗ )p ), where x̂ := (λ +
ϕ)−1 (x). It follows that for all λ > 0 we have
gλ |Sp = g|Sp , gλ (Sp , Tp N ) = 0 and gλ (x ∗ , y ∗ ) = g0 (λϕ(λ + ϕ)−1 (x), y)
for all x, y ∈ m. From this, the first and the fourth assertion on the limits of gλ
as λ → ∞ and λ → 0 follow. The second statement follows since
−1
μλ μλ
μ(λϕ(λ + ϕ)−1 )(μ + (λϕ(λ + ϕ)−1 ))−1 = ϕ +ϕ ,
μ+λ μ+λ
and the third follows since for λ < λ0 we can write λ = μ+λ
μλ0
0
for μ := λλλ 0
0 −λ
>0
and then apply the preceding statement and the fact that a Cheeger deformation
of a nonnegatively curved metric is again nonnegatively curved. 
130 L. J. Schwachhöfer

7.3 Homogeneous metrics of nonnegative curvature

By virtue of Proposition 7.2.2, we should consider Cheeger deformations with
small parameters λ > 0 when searching for nonnegatively curved metrics.
Moreover, since the G-orbits of M are homogeneous spaces, the last statement
of Proposition 7.2.2 implies that the metrics on the orbits approach a biinvariant
metric for Cheeger deformations with small λ > 0. Therefore, it is important to
investigate the question of which G-invariant metrics on a homogeneous space
close to the normal homogeneous one have nonnegative curvature.
In general, this question is too hard to answer. But we can give a partial
answer if G/H is the total space of a homogeneous fibration.
Theorem 7.3.1 ([9]) Let H ⊂ K ⊂ G be compact Lie groups with Lie algebras
h ⊂ k ⊂ g, and consider the homogeneous fibration K/H → G/H → G/K.
Moreover, denote the orthogonal decompositions with respect to some biinvari-
ant metric g0 on g as
g = h ⊕ p = k ⊕ s and k = h ⊕ m (7.1)
(i) The metric on G/H obtained from the normal homogeneous one by shrink-
ing the fibers of this fibration is always nonnegatively curved.
(ii) The metric on G/H obtained from the normal homogeneous one by enlarg-
ing the fibers of this fibration by a factor (1 + ε) is nonnegatively curved
for ε > 0 sufficiently small if and only if there is a constant C > 0 such
that for all X, Y ∈ p,
|[Xm , Y m ]m | ≤ C · |[X, Y ]|, (7.2)
where the superscripts refer to the decomposition (7.1).
In particular, if the fiber K/H is a symmetric space, then [Xm , Y m ]m = 0, so
that in this case, the fiber can always be enlarged while maintaining nonnegative
curvature. All metrics on G/H in this theorem are induced by the left-invariant
metric given by the inner product
gt := tg0 |k + g0 |s (7.3)
on g; indeed, shrinking (respectively, enlarging) the fibers corresponds to the
case t < 1 (respectively, t > 1).
Proof For the first statement, we observe that applying the Cheeger deforma-
tion to the action of K on G by right multiplication, we obtain that gλ is the
left-invariant metric which on g = Te G is given as
λ
gλ = g0 |k + g0 |s ,
1+λ
 Nonnegative curvature on disk bundles 131

hence this metric on G is nonnegatively curved, and so is the induced metric

on G/H , which corresponds to scaling the fibers by t := λ/(1 + λ) ∈ (0, 1).
The second part follows from a more careful investigation of the curvature
formula for invariant metrics on homogeneous spaces which we shall not
present here. The details may be found in [9]. 

If H is the trivial group so that G/H = G, then (7.2) in the above theorem
is equivalent to a simpler criterion. Namely, we have the following

Theorem 7.3.2 ([8]) The left-invariant metric on G induced by (7.3) has

nonnegative curvature for some t > 1 if and only if the semi-simple part of k is
an ideal of g. In this case, the metric has nonnegative curvature for all t ≤ 4/3.

7.4 Collar metrics of nonnegative curvature

Let M be a manifold with boundary N := ∂M. A Riemannian metric g on M
is called a collar metric, if there is a neighborhood of N which is isometric to
([0, ε) × N, dt 2 + gN ) for some ε > 0 and some metric gN on N . Evidently,
if M1 , M2 are manifolds with ∂M1 = ∂M2 =: N , then collar metrics gi on Mi
whose restriction to the boundary N are isometric can be glued together by an
isometry to give a smooth metric on
M := M1 ∪N M2 .
This simple principle has proven very useful to construct metrics of non-
negative curvature, as the following examples due to J. Cheeger illustrate.

Theorem 7.4.1 ([1]) Let M̂ n be a compact rank-one symmetric space (CROSS),

and let M := M̂ n \Bε (p) be the complement of a small open ball. Then there is
a nonnegatively curved collar metric on M such that the metric on ∂M = S n−1
is the round metric.

Corollary 7.4.2 ([1]) Let M1n and M2n be CROSSes of equal dimension. Then
both M1 #M2 and M1 #M 2 admit nonnegatively curved metrics.

The corollary is an immediate consequence of the theorem, since the glueing

in this case corresponds to taking the connected sum, and changing the orien-
tation on one of the summands before glueing is possible. In order to prove the
theorem, Cheeger used the fact that M is always a homogeneous disk bundle
over a CROSS of lower dimension, and he was able to perform the construction
of the metric in a fairly explicit way.
If M is a manifold with boundary N = ∂M which admits a nonnegatively
curved collar metric, then the Soul Theorem ([2]) implies that there is a totally
132 L. J. Schwachhöfer

geodesic submanifold S ⊂ M, called the soul of M, such that M is the total

space of a unit disk bundle D → S, whence N → S is a sphere bundle.
Given a disk bundle M → S, it is thus a reasonable question to ask if M
admits a nonnegatively curved collar metric and if so, what are the possible
restrictions of such a metric to N = ∂M. If M → S carries such a metric, then
evidently it can be extended to a complete nonnegatively curved metric on the
corresponding vector bundle M̂ → S by defining the metric to be the product
([r0 , ∞) × N, dt 2 + gN ) on M̂\M. The converse of this statement also holds:

Theorem 7.4.3 ([6]) Let (M̂, g) be a complete open manifold of nonnegative

curvature, so that by the soul theorem M̂ is the total space of a vector bundle
M̂ → S. Then there is a nonnegatively curved collar metric on the unit disk
bundle M ⊂ M̂.

As an immediate consequence of this result, we may conclude the following.

Corollary 7.4.4 Let M̂k , k = 1, 2 be open manifolds admitting a complete

nonnegatively curved metric, and let Mk ⊂ M̂k be the corresponding unit disk
bundles. Then there is a nonnegatively curved metric on the (closed) manifold

∂(M1 × M2 ).

Proof We may write ∂(M1 × M2 ) = (M1 × ∂M2 ) ∪∂M1 ×∂M2 (∂M1 × M2 ).

Now consider the nonnegatively curved collar metrics gk on Mk whose exis-
tence is guaranteed by Theorem 7.4.3. Then the products of these metrics
induce nonnegatively curved collar metrics on M1 × ∂M2 and ∂M1 × M2
whose boundary is isometric to (∂M1 × ∂M2 , g1 + g2 ), so that these metrics
may be glued together. 

7.5 Bundles with normal homogeneous collar

A special class of vector bundles which we wish to consider are homogeneous
vector bundles and disk bundles. These are bundles of the form

M̂ := G ×K V and M := G ×K BR (0) ⊂ G ×K V , respectively,

where K ⊂ G acts on the Euclidean vector space V by some orthogonal rep-

resentation K → O(V ) which is transitive on the unit sphere S V ⊂ V , so that
S V = K/H for some subgroup H ⊂ K.
Note that M̂ is the quotient of G × V by the free left action of K given by
k % (g, v) := (gk −1 , k · v). Thus, any K-invariant nonnegatively curved metric
on G × V induces a nonnegatively curved metric on M̂, and this metric on
M ⊂ M̂ is a collar metric if the metric on G × BR (0) ⊂ G × V is.
 Nonnegative curvature on disk bundles 133

If follows that each homogeneous disk bundle has G-invariant nonnegatively

curved collar metrics: choose on G × BR (0) the direct sum of a nonnegatively
curved left invariant, AdK -invariant metric on G and a nonnegatively curved K-
invariant collar metric on BR (0), such as a “cigar metric”, i.e., a nonnegatively
curved O(V )-invariant metric which is a cylinder outside a compact set. Clearly,
such metrics exist. However, the induced metric on the boundary G/H is in
general not normal homogeneous, and this limits the possibilities of glueing
two such bundles.
Definition 7.5.1 Let M := G ×K BR (0) ⊂ G ×K V −→ G/K be a homo-
geneous disk bundle with K → O(V ) acting transitively on the sphere
S V = K/H . A metric on M is called a normal homogeneous collar metric
if a neighborhood of the boundary ∂M ∼ = G/H is G-equivariantly isometric to
([0, ε) × G/H, dt 2 + g0 ), where g0 is a normal homogeneous metric on G/H .
Let us suppose that the metric on M is a submersion of (G × BR (0), gg +
gV ) where gg is a left-invariant AdK -invariant metric on G and gV is a K-
invariant collar metric on BR (0) whose boundary metric on K/H = S V ⊂ V
is induced by a left-invariant metric gk on K. Then the metric on the boundary
∂M ∼ = G/H is induced by the Riemannian submersion
(G × K, gg + gk ) −→ G −→ G/H. (7.4)
It is now of interest to see how these metrics on G and BR (0) ⊂ V can be
chosen such that the submersion (7.4) induces a normal homogeneous metric
on G/H and hence a normal homogeneous collar metric on M. For the metric
on BR (0) ⊂ V , we may use the following result.
Theorem 7.5.2 ([11]) Let K → O(V ) be a representation of a Lie group K
on a Euclidean vector space V which is transitive on the unit sphere S V ⊂ V .
Then there is a nonnegatively curved collar metric on BR (0) ⊂ V which is
NormO(V ) K-invariant and is normal homogeneous on S V = K/H .
Let L → O(V ) be a representation which is transitive on the unit sphere
S V ⊂ V , and extend this to a representation K × L → O(V ), i.e., K maps to
the centralizer ZO(V ) (L). Since the normal homomogeneous collar metric on V
from the Theorem 7.5.2 is (K × L)-invariant, there is an induced nonnegatively
curved normal homogeneous collar metric on
(G × L) ×K×L V −→ G/K,
which as a bundle can be regarded as G ×K V → G/K, where K acts on
V as a subgroup of ZO(V ) (L) ⊂ O(V ). Such a homogeneous bundle is called
essentially trivial. Thus, all essentially trivial homogeneous disk bundles carry
134 L. J. Schwachhöfer

a nonnegatively curved collar metric whose boundary is normal homogeneous

with respect to the action of G × L.
The following gives an exhaustive description of all essentially trivial bun-
dles which are not trivial.

Example 7.5.3
(i) Z2 -quotients. Let K  K  ⊂ G be such that K  /K ∼= Z2 , and let L →
O(V ) be a representation which is transitive on the unit sphere. More-
over, define the representation K  → K  /K ∼
= {±I d} ⊂ O(V ). Then the
essentially trivial bundle

(G × L) ×K  ×L V ∼
= G ×K  V −→ G/K 

is the quotient of the trivial bundle G/K × V by Z2 ∼ = K  /K which acts

on G/K from the right and on V as {±I d}.
(ii) Rank-one bundles. If dim V = 1, i.e., V = R, then O(V ) ∼ = Z2 , hence
any homogeneous bundle of rank one is of the form G ×K V → G/K 


with a surjective map K  → O(V ) ∼ = Z2 . The kernel of this map is a

normal subgroup K  K  with K  /K ∼ = Z2 . Thus, this is a special case of
the preceding, i.e., we can write any non-trivial rank one homogeneous
bundle in the form

(G × Z2 ) ×K  ×Z2 V ∼
= G ×K  V → G/K  ,

which is the quotient (G/K × V )/Z2 as above.

(iii) S 1 -quotients. Similarly, if V is a complex hermitean vector space and
L → U (V ) is transitive on the unit sphere, then we consider subgroups
K  K  ⊂ G such that K  /K = S 1 . Then the essentially trivial bundle

(G × L) ×K  ×L V ∼
= G ×K  V −→ G/K 

is the quotient of the trivial bundle G/K × V by the action of S 1 = K  /K

on G/K from the right and on V by scalar multiplication by S 1 ⊂ C.
(iv) Sp(1)-quotients. Likewise, let V be a quaternionic hermitean vector space
and suppose that L → Sp(V ) is surjective and hence transitive on the unit
sphere. Again, consider subgroups K  K  ⊂ G such that K  /K = Sp(1).
Then the essentially trivial bundle

(G × L) ×K  ×L V ∼
= G ×K  V −→ G/K 

is the quotient of the trivial bundle G/K × V by the action of Sp(1) =

K  /K on G/K from the right and on V by scalar multiplication by
Sp(1) ⊂ H.
 Nonnegative curvature on disk bundles 135

Thus, all these bundles admit nonnegatively curved normal homogeneous

collar metrics. Note that this description of essentially trivial bundles is exhaus-
tive, since by the classification of transitive actions of a Lie group L on a sphere
S V ⊂ V the centralizer ZO(V ) (L) ⊂ O(V ) is either trivial or one of Z2 , S 1 or
Sp(1).
By virtue of Theorem 7.5.2, we may assume that gk in (7.4) is a biinvariant
metric. If we choose gg as a biinvariant metric such that gk = (gg )|k , then it
follows that the induced metric on G/H is obtained by shrinking the fiber of the
normal homogeneous metric of the fibration S V = K/H → G/H → G/K.
In particular, this metric on the boundary G/H is not normal homogeneous.
In order to achieve a normal homogeneous boundary metric on G/H , we
have to define the metric on g = k ⊕ s by

gg := (1 + ε)(g0 )|k + (g0 )|s , (7.5)

where g0 is a biinvariant metric. When choosing gk := (1 + ε−1 )(g0 )|k , then

(7.4) induces the metric g0 on G and hence a normal homogeneous metric on
G/H .
Recall from Theorem 7.3.2 that the metric gg from (7.5) has nonnegative
curvature for some ε > 0 if and only if the semi-simple part of k is an ideal
of g. In particular, this is satisfied if dim(V ) ≤ 2 in which case k is at most
one-dimensional and hence abelian. Therefore, we obtain the following result.

Theorem 7.5.4 ([5]) Let M = G ×K V be a homogeneous disk bundle of

rank dim V ≤ 2. Then M admits a nonnegatively curved normal homogeneous
collar metric.

However, if dim V ≥ 3 and the bundle M → G/K is not essentially trivial,

then the semi-simple part of k is not an ideal. Thus, by Theorem 7.3.2, the
left-invariant metric gg from (7.5) has some negative curvature on G. But in
order for the submersion metric on M = G ×K V to be nonnegatively curved,
it suffices that gg has nonnegative curvature on all planes which are contained
in p. Some interesting examples where this occurs have been described in
[9]. Once again, the proof of this theorem involves a direct calculation of the
curvature formula for homogeneous metrics.

Theorem 7.5.5 ([9]) Let H ⊂ K ⊂ G be compact Lie groups with Lie alge-
bras h ⊂ k ⊂ g, and consider the decompositions from (7.1). If there exists
C > 0 such that for all X = Xm + Xs , Y = Y m + Y s ∈ m ⊕ s = p we have

|Xm ∧ Y m | ≤ C · |[X, Y ]|, (7.6)

136 L. J. Schwachhöfer

then any left-invariant metric on G sufficiently close to g0 which is AdH -

invariant and is a constant multiple of g0 on s and h (but arbitrary on m) has
nonnegative sectional curvature on all planes contained in p.

That is, for all Lie groups H ⊂ K ⊂ G satisfying this criterion, the metric
gg from (7.5) has nonnegative curvature on all planes contained in p for small
ε > 0, hence if K/H = S V , then the corresponding homogeneous vector bun-
dle G ×K V → G/K has a nonnegatively curved normal homogeneous collar
metric.
As it turns out, condition (7.6) is almost necessary and sufficient for a
homogeneous disk bundle to admit such a metric. Namely, we have

Theorem 7.5.6 ([10]) Let G ×K V → G/K be a homogeneous vector bundle

which admits a nonnegatively curved normal homogeneous collar metric, let
H ⊂ K be the subgroup for which K/H = S V , and decompose the Lie algebras
h ⊂ k ⊂ g as in (7.1).
If m1 ⊂ m is any non-trivial AdH -irreducible subspace such that m contains
no other AdH -irreducible factor equivalent to m1 , then there exists a constant
C > 0 such that, for all X = Xm + Xs , Y = Y m + Y s ∈ m1 ⊕ s, condition
(7.6) holds.

When K/H is isotropy irreducible, the choice m1 = m yields a converse

to Theorem 7.5.5. The proof of this theorem uses Perelman’s result on the
existence of parallel vector fields on a totally convex set ([7]). This implies that
along a normal geodesic c(t) the metric on the principal orbit through c(t) is
induced by a metric of the form gt = (g0 )|s + g̃t |k . Then one shows that such
a metric must have some negative curvature close to the collar unless (7.6) is
satisfied.
With these two results, one can give an almost complete classification of
homogeneous disk bundles which admit nonnegatively curved normal homo-
geneous collar metrics.

Theorem 7.5.7 ([10]) Let M := G ×K V → G/K be a cohomogeneity one

homogeneous vector bundle which is not essentially trivial. If M admits a
G-invariant normal homogeneous collar metric of nonnegative curvature, then
the rank of this bundle must be in the set {2, 3, 4, 6, 8}.

By Theorem 7.5.4, all rank-two bundles admit a G-invariant normal homo-

geneous collar metric of nonnegative curvature. In the higher rank case, the
situation is much more restricted. For rank eight, we have the following com-
plete classification.
 Nonnegative curvature on disk bundles 137

Theorem 7.5.8 ([10]) Let M := G ×K V → G/K be a G-irreducible coho-

mogeneity one homogeneous vector bundle which is not essentially trivial and
such that dimR V = 8. Then M admits a G-invariant normal homogeneous col-
lar metric of nonnegative curvature if and only if M is finitely G-equivariantly
covered by one of the following:
(i) Spin(p + 9) ×Spin(8) R8 for p ∈ {0, 1, 2}, where Spin(8) acts on R8 by a
spin representation, and Spin(8) ⊂ Spin(p + 9) is the lift of the standard
inclusion SO(8) ⊂ SO(p + 9).
(ii) Spin(p + 8) ×Spin(7) R8 for p ∈ {0, 1}, where Spin(7) acts on R8 by the
spin representation, and Spin(7) ⊂ Spin(p + 8) is the lift of the standard
inclusion SO(7) ⊂ SO(p + 8).
(iii) Spin(7) ×Spin(6) C4 , with the standard representation of Spin(6) ∼ = SU (4)
on C4 .
(iv) A quotient of one of the preceding examples:
1 (Spin(p + 9) · G ) ×Spin(8)·H  R8 for p ∈ {1, 2} and an arbitrary com-
pact Lie group G and H  ⊂ Spin(p + 1) · G with H  ⊂ G , which
acts trivially on R8 .
2 (Spin(9) · G ) ×Spin(7)·H  R8 for an arbitrary compact Lie group G and
H  ⊂ Spin(2) · G = S 1 · G with H  ⊂ G , which acts trivially on R8 .
3 (Spin(7) · G ) ×Spin(6)·S 1 ·H  C4 for an arbitrary compact Lie group G ⊃
S 1 · H  , where S 1 ⊂ G acts on C4 by multiples of the identity, and
H  ⊂ G acts trivially.
Here, for Lie groups L1 , L2 , we denote by L1 · L2 the quotient of L1 × L2
by a finite subgroup of the center. The term G-irreducible means that M is not
G-equivariantly finitely covered by a bundle of the form (G1 /H1 ) × M  with
dim(G1 /H1 ) > 0 and M  a cohomogeneity one homogeneous vector bundle.
This hypothesis is natural because, for G-reducible bundles, our problem easily
reduces to deciding whether M  admits such a metric.
For rank 6 bundles, we obtain the following
Theorem 7.5.9 ([10]) Let M := G ×K V → G/K be a G-irreducible coho-
mogeneity one homogeneous vector bundle which is not essentially trivial and
such that dimR V = 6. Then M admits a G-invariant normal homogeneous col-
lar metric of nonnegative curvature if and only if M is finitely G-equivariantly
covered by one of the following:
(i) SU (5) ×SU (4) R6 , with the irreducible action of SU (4) ∼
= Spin(6) on R6 .
(ii) (SU (5) · G ) ×SU (4)·H  R for an arbitrary compact Lie group G and H  ⊂
 6

S 1 · G with S 1 ⊂ SU (5) being the centralizer of SU (4), and H  ⊂ G acts

trivially on R6 .
138 L. J. Schwachhöfer

To describe our results for rank three and four bundles, we require some
notation for subgroups of the exceptional Lie group G2 . Let SO(4) ⊂ G2
and SU (3) ⊂ G2 be the isotropy groups of the symmetric space G2 /SO(4)
and the sphere S 6 = G2 /SU (3), respectively. After conjugating these groups
appropriately, their intersection can be made isomorphic to U (2), and we
let SU (2)1 ⊂ SO(4) ∩ SU (3) ⊂ G2 be the simple part of this intersection.
Note that SU (2)1 ⊂ SO(4) is normal, and we denote its centralizer in G2
by SU (2)3 ⊂ SO(4) ⊂ G2 . (The subscripts of the SU (2)-subgroups of SO(4)
denote their maximal weight for the isotropy representation of G2 /SO(4).)
Using this notation, we can make the following statement about the rank-three
case.

Theorem 7.5.10 ([10]) Let M := G ×K V be a G-irreducible cohomogene-

ity one homogeneous vector bundle which is not essentially trivial such that
dimR V = 3. If M admits a nonnegatively curved G-invariant normal homo-
geneous collar metric, then M must be finitely G-equivariantly covered by one
of the following.

(i) M1 = G2 ×SO(4) su(2)3 , where SO(4) acts on su(2)3  so(4) by the adjoint
representation.
(ii) M2 = (Sp(p + 1) · G ) ×Sp(1)·H  sp(1) with H  ⊂ Sp(p) · G , where
Sp(1) · H  acts on sp(1)  sp(1) ⊕ h by the adjoint representation.

Further, M1 admits such a metric, as does M2 with G = 1 and H  = Sp(p).

Finally, in the rank-four case, we have the following examples, which are
all related to those in Theorem 7.5.10.

Theorem 7.5.11 ([10]) The following cohomogeneity one homogeneous vector

bundles (orbifold bundles, respectively) of rank four admit G-invariant normal
homogeneous collar metrics of nonnegative curvature:

(i) G2 ×SO(4) (H/± 1), where SU (2)1 ⊂ SO(4) acts trivially and SU (2)3 ⊂
SO(4) by left multiplication on H/± 1. Note that this is an orbifold bundle
only.
(ii) (G2 × G ) ×SO(4)×SU (2) (H/± 1), where SU (2)1 ⊂ SO(4) acts trivially
and SU (2)3 ⊂ SO(4) by left multiplication, whereas SU (2) ⊂ G acts
by right multiplication on H/± 1 with G arbitrary. Note that these are
orbifold bundles only.
(iii) Sp(p + 1) ×Sp(1)×Sp(p) H where Sp(p) acts trivially and Sp(1) by left
multiplication on H.
 Nonnegative curvature on disk bundles 139

(iv) (Sp(p + 1) × G ) ×Sp(1)×Sp(p)×Sp(1) H where Sp(p) acts trivially and

Sp(1) by left multiplication, whereas Sp(1) ⊂ G acts by right multi-
plication on H with G arbitrary.

7.6 Cohomogeneity one manifolds

A cohomogeneity one manifold is a manifold M with an action of a compact
Lie group G such that dim(M/G) = 1. For a cohomogeneity one manifold,
there are four possibilities for the quotient space.

(i) M/G ∼ = R. In this case, M is equivariantly diffeomorphic to R × G/H

for some subgroup H ⊂ G. In particular, M admits a G-invariant metric
of nonnegative curvature.
(ii) M/G ∼ = [0, ∞). In this case, M has one non-principal orbit and is a
homogeneous vector bundle as described in the preceding section. In
particular, M admits a G-invariant metric of nonnegative curvature.
(iii) M/G ∼ = S 1 . Then M is finitely covered by S 1 × G/H for some subgroup
H ⊂ G. Again, M admits a G-invariant metric of nonnegative curvature.
(iv) M/G ∼ = [−1, 1]. In this case, M has two non-principal orbits and can be
written as

M = D− ∪G/H D+ ,

where D± = G ×K± V± are homogeneous disk bundles with common

boundary G/H . That is, H ⊂ K± and K± /H is the unit sphere S V± ⊂ V± .

It is the last of these classes in which we are interested, as it is obtained by

glueing two homogeneous disk bundles together along their common boundary
which is a principal orbit. Conversely, given Lie groups H ⊂ {K+ , K− } ⊂ G
such that K± /H is a sphere, then there is a unique cohomogeneity one manifold
which is determined by these groups. Thus, these groups can be used to identify
the manifold.
As it turns out, not all cohomogeneity one manifolds with two singular orbits
admit invariant metrics of nonnegative curvature ([3]). However, we have the
following result.

Theorem 7.6.1 ([11]) Let M be a closed cohomogeneity one manifold. Then

M admits G-invariant metrics of almost nonnegative curvature, i.e., there is a
sequence of G-invariant metrics gn on M such that diam(M, gn )2 sec(M, gn ) >
−1/n.
140 L. J. Schwachhöfer

This is proven once again by using Cheeger deformations: Let gM be any

G-invariant metric on M, and let g0 be a biinvariant metric on G. Then the
metric gλ on M induced by the submersion (G × M, λg0 + gM ) → (M, gλ )
has uniformly bounded diameter for small λ > 0, and one verifies that the
lower curvature bound of gλ tends to 0 as λ → 0.
From the results in the preceding section, we see that we obtain a nonnega-
tively curved metric on a cohomogeneity one manifold with two singular orbits
if the corresponding two disk bundles admits nonnegatively curved normal
homogeneous collar metrics. Thus, Theorem 7.5.4 immediately implies

Theorem 7.6.2 ([5]) Let M be a cohomogeneity one manifold with two non-
principal orbits of codimension at most 2. Then M admits a G-invariant metric
of nonnegative curvature.

This result yields some spectacular examples of manifolds which admit non-
negatively curved metrics, such as all four (unoriented) diffeomorphism types
of RP5 and ten out of the fourteen (unoriented) exotic spheres of dimension 7.
Unfortunately, the remaining homogeneous disk bundles admitting nonneg-
atively curved normal homogeneous collar metrics do not yield new examples
of nonnegatively curved cohomogeneity one manifolds, since on glueing two
of them, we obtain either the restriction of a homogeneous action on M, or
there is a Riemannian submersion M̂ → M, where M̂ is known to have a
nonnegatively curved metric.
For instance, by glueing together two disk bundles of the second type of
Theorem 7.5.8 with p = 0, after applying outer automorphisms of Spin(8), we
conclude that the primitive cohomogeneity one manifold given by the group
diagram G2 ⊂ {Spin+ (7), Spin− (7)} ⊂ Spin(8) admits a metric of nonneg-
ative curvature with a totally geodesic normal homogeneous principal orbit.
However, this manifold is diffeomorphic to the sphere S 15 ([4]).

References
[1] J. Cheeger, Some examples of manifolds of nonnegative curvature, J. Diff. Geom.
8 (1973) 623–628
[2] J. Cheeger, D. Gromoll, On the structure of complete manifolds of nonnegative
curvature, Ann. of Math. 96 (1972) 413–443.
[3] K. Grove, L. Verdiani, B. Wilking, W. Ziller, Nonnegative curvature obstructions
in cohomogeneity one and the Kervaire spheres, Ann. Sc. Norm. Super. Pisa, Cl.
Sci. (5) 5, No. 2 (2006) 159–170
[4] K. Grove, B. Wilking, W. Ziller, Positively curved cohomogeneity one manifolds
and 3-Sasakian geometry, J. Diff. Geom. 78 (2008) 33–111.
 Nonnegative curvature on disk bundles 141

[5] K. Grove, W. Ziller, Curvature and symmetry of Milnor spheres, Ann. of Math.
(2) 152 (2000) 331–367.
[6] L. Guijarro, Improving the metric in an open manifold with nonnegative curvature,
Proc. Amer. Math. Soc., 126 (1998) 1541–1545
[7] G. Perelman, Proof of the soul conjecture of Cheeger and Gromoll, J. Diff. Geom.
40 (1994), 209–212.
[8] L. Schwachhöfer, A remark on left invariant metrics on compact Lie groups,
Arch.Math. 90 (2008) 158–162
[9] L. Schwachhöfer, K. Tapp, Homogeneous Metrics with nonnegative curvature,
Jour. Geom. Anal. 19 (2009) 929–943
[10] L. Schwachhöfer, K. Tapp, Cohomogeneity one disk bundles with normal homo-
geneous collars, Proc. London Math. Soc. 99, (2009) 609–632
[11] L. Schwachhöfer and W. Tuschmann, Almost nonnegative curvature and coho-
mogeneity one, Preprint no. 62/2001, Max-Planck-Institut für Mathematik in den
Naturwissenschaften Leipzig, http://www.mis.mpg.de/cgi-bin/preprints.pl
[12] W. Ziller, Examples of Riemannian manifolds with nonnegative sectional curva-
ture, Metric and Comparison Geometry, Surv. Diff. Geom. 11, ed. K. Grove and
J. Cheeger, Intern. Press, 2007

Author’s address:

Fakultät für Mathematik,

Technische Universität Dortmund,
Vogelpothsweg 87,
44221 Dortmund,
Germany
[email protected]
 8
Morse theory and stable pairs
richard a. wentworth and graeme wilkin

Abstract
We study the Morse theory of the Yang-Mills-Higgs functional on the space
of pairs (A, #), where A is a unitary connection on a rank 2 hermitian vector
bundle over a compact Riemann surface, and # is a holomorphic section
of (E, dA ). We prove that a certain explicitly defined substratification of the
Morse stratification is perfect in the sense of G-equivariant cohomology, where
G denotes the unitary gauge group. As a consequence, Kirwan surjectivity holds
for pairs. It also follows that the twist embedding into higher degree induces
a surjection on equivariant cohomology. This may be interpreted as a rank 2
version of the analogous statement for symmetric products of Riemann surfaces.
Finally, we compute the G-equivariant Poincaré polynomial of the space of τ -
semistable pairs. In particular, we recover an earlier result of Thaddeus. The
analysis provides an interpretation of the Thaddeus flips in terms of a variation
of Morse functions.

8.1 Introduction
In this paper we revisit the notion of a stable pair on a Riemann surface. We
introduce new techniques for the computation of the equivariant cohomology
of moduli spaces. The main ingredient is a version of Morse theory in the spirit
of Atiyah and Bott [1] adapted to the singular infinite dimensional space of
holomorphic pairs.
Recall first the basic idea. Let E be a hermitian vector bundle over a closed
Riemann surface M of genus g ≥ 2. The space A(E) of unitary connections on
E is an infinite dimensional affine space with an action of the group G of unitary
gauge transformations. Via the Chern connection there is an isomorphism
A → dA between A(E) and the space of (integrable) Dolbeault operators (i.e.

142
 Morse theory and stable pairs 143

holomorphic structures) on E. One of the key observations of Atiyah-Bott is

that the Morse theory of a suitable G-invariant functional on A(E), namely
the Yang-Mills functional, gives rise to a smooth stratification (see also [6]).
Moreover, this stratification is G-equivariantly perfect in the sense that the long
exact sequences for the equivariant cohomology of successive pairs split. Since
A(E) is contractible, this gives an effective method, inductive on the rank of
E, for computing the equivariant cohomology of the minimum, which consists
of projectively flat connections.
Consider now a configuration space B(E) consisting of pairs (A, #), where
A ∈ A(E) and # is a section of a vector bundle associated to E. We impose
the condition that # be dA -holomorphic. Note that B(E) is still contractible,
since an equivariant retraction of B(E) to A(E) is given by simply scaling #.
It is therefore reasonable to attempt an inductive computation of equivariant
cohomology as above. A problem arises, however, from the singularities caused
by jumps in the dimension of the kernel as A varies. Nevertheless, the methods
introduced in [8] for the case of Higgs bundles demonstrate that in certain cases
this difficulty can be managed.
Below we apply this approach to the moduli space of rank 2, degree d, τ -
semistable pairs Mτ,d = Bss τ
(E)//G C introduced by Bradlow [3] and Bradlow-
Daskalopoulos [4]. In this case, # is a holomorphic section of E, and the
Yang-Mills functional YM(A) is replaced by the Yang-Mills-Higgs functional
YMH(A, #). We give a description of the algebraic and Morse theoretic strati-
fications of B(E). These stratifications, as well as the moduli space, depend on
a real parameter τ , and since Mτ,d is nonempty only for d/2 ≤ τ ≤ d, we shall
always assume this bound for τ . For generic τ , G acts freely, and the quotient
is geometric.
We will see that, as in [6, 7, 8], the algebraic and Morse stratifications agree
(see Theorem 8.3.9). Because of singularities, however, the Morse stratification
actually fails to be perfect in this case. We identify precisely how this comes
about, and in fact we will show that this “failure of perfection” exactly cancels
between different strata, so that there is a substratification that is indeed perfect
(see Theorem 8.3.11). We formulate this result as

Theorem 8.1.1 (Equivariantly perfect stratification) For every τ , d/2 < τ <
d, there is a G-invariant stratification of B(E) defined via the Yang-Mills-Higgs
flow that is perfect in G-equivariant cohomology.

The fact that perfection fails for the Morse stratification but holds for a
substratification seems to be a new phenomenon. In any case, as with vector
bundles, Theorem 8.1.1 allows us to compute the G-equivariant cohomology
144 R. A. Wentworth and G. Wilkin

of the open stratum Bssτ

(E). Explicit formulas in terms of symmetric products
of M are given in Theorems 8.4.1 and 8.4.2.
There is a natural map (called the Kirwan map) from the cohomology of
the classifying space BG of G to the equivariant cohomology of the stratum
of τ -semistable pairs Bss
τ
(E) ⊂ B(E), coming from inclusion (see [13]). One
of the consequences of the work of Atiyah-Bott is that the analogous map is
surjective for the case of semistable bundles. The same is true for pairs:
Theorem 8.1.2 (Kirwan surjectivity) The Kirwan map H ∗ (BG) →
HG∗ (Bss
τ
(E)) is surjective. In particular, for generic τ , H ∗ (BG) → H ∗ (Mτ,d )
is surjective.
As noted above, for non-integer values of τ , d/2 < τ < d, Mτ,d is a smooth
projective algebraic manifold of dimension d + 2g − 2, and the equivariant
cohomology of Bss τ
(E) is identical to the ordinary cohomology of Mτ,d . The
computation of equivariant cohomology presented here then recovers the result
of Thaddeus in [20], who computed the cohomology using different methods.
Namely, he gives an explicit description of the modifications, or “flips”, in
Mτ,d as the parameter τ varies. At integer values there is a change in stability
conditions. Below, we show how the change in cohomology arising from a
flip may be reinterpreted as a variation of the Morse function. This is perhaps
not surprising in view of the construction in [5]. However, here we work
directly on the infinite dimensional space. The basic idea is that there is a
one parameter choice of Morse functions fτ on B. The minimum fτ−1 (0)/G 
Mτ,d , and the cohomology of Mτ,d may, in principle, be computed from the
cohomology of the higher critical sets. As τ varies past certain critical values,
new critical sets are created while others merge. Moreover, indices of critical
sets can jump. All this taken together accounts for the change in topology of the
minimum.
There are several important points in this interpretation. One is that the
subvarieties responsible for the change in cohomology observed by Thaddeus
as the parameter varies are somehow directly built into the Morse theory,
even for a fixed τ , in the guise of higher critical sets. This example also
exhibits computations at critical strata that can be carried out in the presence of
singular normal cusps, as opposed to the singular normal vector bundles in [8].
These ideas may be useful for computations in higher rank or for other moduli
spaces.
The critical set corresponding to minimal Yang-Mills connections, regarded
as a subset of B(E) by setting # ≡ 0, is special from the point of view of
the Morse theory. In particular, essentially because of issues regarding Brill-
Noether loci in the moduli space of vector bundles, we can only directly
 Morse theory and stable pairs 145

prove the perfection of the stratification at this step, and the crucial Morse-
Bott lemma (Theorem 8.3.18), for d > 4g − 4. This we do in Section 8.3.5.
By contrast, for the other critical strata there is no such requirement on the
degree. Using this fact, we then give an inductive argument by twisting E
by a positive line bundle and embedding B(E) into the space of pairs for
higher degree, thus indirectly concluding the splitting of the associated long
exact sequence even at minimal Yang-Mills connections in low degree (see
Section 8.3.7).
This line of reasoning leads to another interesting consequence. For τ close
to d/2, there is a surjective holomorphic map from Mτ,d to the moduli space
of semistable rank 2 bundles of degree d. This is the rank 2 version of the
Abel-Jacobi map [4]. In this sense, Mτ,d is a generalization of the d-th sym-
metric product S d M of M. Choosing an effective divisor on M of degree k,
there is a natural inclusion S d M → S d+k M, and it was shown by MacDonald
in (14.3) of [16] that this inclusion induces a surjection on rational cohomol-
ogy. A similar construction works for rank 2 pairs, except now d → d + 2k,
while there is also a shift in the parameter τ → τ + k. We will prove the
following.

Theorem 8.1.3 (Embedding in higher degree) Let deg E = d and deg E =

τ +k 
d + 2k. Then for all d/2 < τ < d, the inclusion Bss (E) → Bss (E) described
τ

above induces a surjection on rational G-equivariant cohomology. In partic-

ular, for generic τ , the inclusion Mτ,d → Mτ +k,d+2k induces a surjection on
rational cohomology.

Remark 8.1.4 It is also possible to construct a moduli space of pairs for which
the isomorphism class of det E is fixed, indeed this is the space studied by
Thaddeus in [20]. The explicit calculations in this paper are all done for the
non-fixed determinant case, however it is worth pointing out here that the
idea is essentially the same for the fixed determinant case, and that the only
major difference between the two cases is in the topology of the critical sets.
In particular, the indexing set τ,d for the stratification is the same in both
cases, and Theorems 8.1.1 and 8.1.2 hold for the fixed determinant spaces as
well.

Acknowledgements The authors warmly thank the organizers of the Leeds

conference, and especially Roger Bielawski, for putting together this volume.
Thanks also to George Daskalopoulos for many discussions, and to the MPIM-
Bonn, where some of the work on this paper was completed.
R.W. supported in part by NSF grant DMS-1037094
146 R. A. Wentworth and G. Wilkin

8.2 Stable pairs

8.2.1 The Harder-Narasimhan stratification
Throughout this paper, E will denote a rank 2 hermitian vector bundle on M
of positive degree d = deg E. We will regard E as a smooth complex vector
bundle, and when endowed with a holomorphic structure that is understood,
we will use the same notation for the holomorphic bundle.
Recall that a holomorphic bundle E of degree d is stable (resp. semistable) if
deg L < d/2 (resp. deg L ≤ d/2) for all holomorphic line subbundles L ⊂ E.

Definition 8.2.1 For a stable holomorphic bundle E, set μ+ (E) = d/2. For E
unstable, let

μ+ (E) = sup{deg L : L ⊂ E a holomorphic line subbundle}

For a holomorphic section # ≡ 0 of E, define deg # to be the number of zeros

of #, counted with multiplicity. Finally, for a holomorphic pair (E, #) let
6
d − deg # # ≡ 0
μ− (E, #) =
d − μ+ (E) # ≡ 0

Definition 8.2.2 ([3]) Given τ , a holomorphic pair (E, #) is called τ -stable

(resp. τ -semistable) if

μ+ (E) < τ < μ− (E, #) (resp. μ+ (E) ≤ τ ≤ μ− (E, #))

As with holomorphic bundles, there is a notion of s-equivalence of strictly

semistable objects. The set Mτ,d of isomorphism classes of semistable pairs,
modulo s-equivalence, has the structure of a projective variety. Note that Mτ,d
is empty if τ ∈ [d/2, d]. For non-integer values of τ ∈ (d/2, d), semistable is
equivalent to stable, and Mτ,d is smooth.
Let A = A(E) denote the infinite dimensional affine space of holomorphic
structures on E, G the group of unitary gauge transformations, and G C its
complexification. The space A may be identified with Dolbeault operators A →
dA : 0 (E) → 0,1 (E), with the inverse of dA given by the Chern connection
with respect to the fixed hermitian structure. When we want to emphasize the
holomorphic bundle, we write (E, dA ).
 
B = B(E) = (A, #) ∈ A × 0 (E) : dA # = 0 (8.1)

Let
 
Bss
τ
= (A, #) ∈ B : ((E, dA ), #) is τ -semistable
 Morse theory and stable pairs 147

Then Mτ,d = Bss τ

//G C , where the double slash identifies s-equivalent orbits.
For generic values of τ , semistability implies stability and G acts freely, and so
this is a geometric quotient.
We now describe the stratification of B associated to the Harder-Narasimhan
filtration, which has an important relationship to the Morse theory picture
that will be discussed below in Section 8.3.2. In the case of rank 2 bundles,
this stratification is particularly easy to describe. For convenience, throughout
this section we fix a generic τ , d/2 < τ < d (it suffices to assume 4τ ∈ Z).
Genericity is used only to give a simple description of the strata in terms of δ.
The extension to special values of τ is straightforward (see Remark 8.2.11).
Note that stability of the pair fails if either of the inequalities in Definition
8.2.2 fails. The two inequalities are not quite independent, but there are some
cases where only one fails and others where both fail. If the latter, it seems
natural to filter by the most destabilizing of the two. With this in mind, we make
the following.
Definition 8.2.3 For a holomorphic pair (E, #), let
δ(E, #) = max {τ − μ− (E, #), μ+ (E) − τ, 0}
Note that δ takes on a discrete and infinite set of nonnegative real values, and
is upper semi-continuous, since both μ+ and −μ− are (observe that deg # ≤
μ+ (E)). We denote the ordered set of such δ by τ,d . Clearly, δ is an integer
modulo ±τ , or δ = τ − d/2. Because of the genericity of τ , the former two
possibilities are mutually exclusive:
Lemma 8.2.4 There is a disjoint union τ,d \ {0} = + −
τ,d ∪ τ,d , with

δ ∈ +
τ,d ⇐⇒ δ = τ − μ− (E, #), for some pair (E, #)
δ ∈ −
τ,d ⇐⇒ δ = μ+ (E) − τ, for some pair (E, #)

Lemma 8.2.5 Suppose (E, #) ∈ Bss

τ
, # ≡ 0. Then
(i) if deg # ≥ d/2, δ(E, #) = μ+ (E) − d + τ .
(ii) if d − τ ≤ deg # < d/2, δ(E, #) = deg # − d + τ .
(iii) if 0 ≤ deg # < d − τ , δ(E, #) = μ+ (E) − τ .
If # ≡ 0, then δ(E, #) = μ+ (E) − d + τ .
Proof If deg # ≥ d/2, then the line subbundle generated by # is the max-
imal destabilizing subbundle of E. Hence, μ+ (E) = deg #, μ− (E, #) =
d − μ+ (E), and so (i) follows from the fact that τ > d/2. For (ii), consider
the extension 0 → L1 → E → L2 → 0, where # ∈ H 0 (L1 ). Then deg L2 =
d − deg #, so μ+ (E, #) ≤ d − deg #. It follows that μ+ (E) − τ ≤ 0. For part
148 R. A. Wentworth and G. Wilkin

(iii), 0 ≤ deg # < d − τ implies τ < μ− (E, #). The last statement is clear,
since τ > d/2 implies τ − μ− (E, #) = μ+ (E) − d + τ > μ+ (E) − τ . 

Corollary 8.2.6 −
τ,d ⊂ (0, d − τ ].

Proof Indeed, if (E, #) is unstable and δ(E, #) = μ+ (E) − τ , then by (3) it

must be that E is unstable and # ≡ 0. From the Harder-Narasimhan filtration
(cf. [14]) 0 → L2 → E → L1 → 0, the projection of # to L1 must also be
nonzero, since deg # < deg L2 . Hence, deg L1 = d − μ+ (E) ≥ 0, and so d −
τ ≥ δ(E, #). 

Remark 8.2.7 If δ ∈ + τ,d and δ < τ − d/2, then δ ≤ τ − d/2 − 1/2. Indeed,
if δ + d − τ = k ∈ Z, the condition forces k < d/2; hence, k ≤ d/2 − 1/2.

Let Iτ,d = [τ − d/2, 2τ − d). We are ready to describe the τ -Harder-

Narasimhan stratification. First, for j > d/2, let Aj ⊂ A be the set of holo-
morphic bundles E of Harder-Narasimhan type μ+ (E) = j . We also set
Ad/2 = Ass . There is an obvious inclusion Aj ⊂ B : A → (A, 0).

(0) δ = 0: The open stratum B0τ = Bssτ

consists of τ -semistable pairs.
+
(Ia ) δ ∈ τ,d ∩ Iτ,d : Then we include the strata Aδ+d−τ . Note that this
includes the semistable stratum Ass . The bundles in this strata that are
not semistable have a unique description as extensions

0 −→ L1 −→ E −→ L2 −→ 0 (8.2)

where deg L1 = μ+ (E) = δ + d − τ .

(Ib ) δ ∈ +τ,d ∩ [2τ − d, +∞): Then Bδ = {(E, #) : μ+ (E) = δ + d − τ }.
τ

These are extensions (8.2), deg L1 = μ+ (E) = δ + d − τ , # ⊂ H 0 (L1 ).

(II+ ) δ ∈ + τ,d ∩ (0, 2τ − d): Then Bδτ = {(E, #) : deg # = δ + d − τ }.
These are extensions (8.2), deg L1 = δ + d − τ , # ⊂ H 0 (L1 ).
(II ) δ ∈ −
−
τ,d : Then Bδ = {(E, #) : μ+ (E) = δ + τ , deg # < d/2}. These
τ

are extensions

0 −→ L2 −→ E −→ L1 −→ 0

where deg L2 = μ+ (E), and the projection of # to H 0 (L1 ) is nonzero.

For simplicity of notation, when τ is fixed we will mostly omit the superscript:
Bδ = Bδτ .

Remark 8.2.8 It is simple to verify that the stratification obtained above coin-
cides with the possible Harder-Narasimhan filtrations of pairs (E, #) consid-
ered as coherent systems (see [15, 18, 12]).
 Morse theory and stable pairs 149

It will be convenient to organize τ,d by the slope of the subbundle in the

maximal destabilizing subpair. Define j : τ,d \ {0} → {d/2} ∪ {k ∈ Z : k ≥
d − τ } by
6
δ + d − τ, δ ∈ +
j (δ) = τ,d
(8.3)
δ + τ, δ ∈ −τ,d

Notice that j (δ) = deg L1 for δ ∈ + −

τ,d , and j (δ) = deg L2 for δ ∈ τ,d , where
L1 , L2 refer to the line subbundles of E in the filtrations above. Note that j is
surjective. It is precisely 2-1 on the image of − τ,d and 1-1 elsewhere (if d odd;
otherwise d/2 labels both the stratum Ass and the strictly semistable bundles
of type II+ ). It is not order preserving but is, of course, order preserving on
each of ± τ,d separately.

Definition 8.2.9 For δ ∈ τ,d , let

, ,
Xδ = Bδ ∪ Aj (δ )
δ  ≤δ , δ  ∈τ,d δ  ≤δ , δ  ∈+
τ,d ∩Iτ,d

For δ ∈ +
τ,d ∩ Iτ,d , let
, ,
Xδ = Bδ ∪ Aj (δ )
δ  ≤δ , δ  ∈τ,d δ  <δ , δ  ∈+
τ,d ∩Iτ,d

For δ ∈ +
τ,d ∩ Iτ,d , let
, ,
Xδ = Bδ ∪ Aj (δ )
δ  <δ , δ  ∈ τ,d δ  <δ , δ  ∈+
τ,d ∩Iτ,d

We call the collection {Xδ , Xδ }δ∈τ,d the τ -Harder-Narasimhan stratification

of B.

Note that Xδ1 ⊂ Xδ  Xδ ⊂ Xδ 2 , where δ1 is the predecessor and δ2 is the
successor of δ in τ,d . If δ ∈ +  
τ,d ∩ Iτ,d , then Xδ = Xδ1 and Xδ = Xδ2 . In the
special case δ = τ − d/2, we have

Xτ −d/2 = Xτ −d/2 ∪ Ass (8.4)

6
 Xδ1 if d is odd
Xτ −d/2 = (8.5)
Xδ1 ∪ Bτ −d/2 if d is even

The following is clear.

150 R. A. Wentworth and G. Wilkin

Figure 8.1 A “flip”

Proposition 8.2.10 The sets {Xδ , Xδ }δ∈τ,d , are locally closed in B, G-
invariant, and satisfy
, ,
B= Xδ = Xδ
δ∈τ,d δ∈τ,d
, ,
Bδ ⊂ Bδ = Bδ ∪ Bδ 
δ≤δ  , δ  ∈ τ,d δ<δ  , δ  ∈τ,d
 , ,
Bδ ⊂ Bδ  = Bδ ∪ Bδ
δ≤δ  , δ  ∈τ,d δ<δ  , δ  ∈τ,d

Remark 8.2.11 To extend this stratification in the case of nongeneric τ , we

define the sets ± + −
τ,d and the corresponding strata as above. For δ ∈ τ,d ∩ τ,d
there are two or possibly three components with the same label.

Let us note the following behavior as τ varies. For τ1 ≤ τ2 , there is a

well-defined map τ1 ,d → τ2 ,d given by δ → max{δ ± (τ2 − τ1 ), 0}, where
± depends on δ ∈ ± +
τ,d . Hence, elements of τ,d (white circles in Figure 8.1
−
below) “move” to the right, and elements of τ,d (dark circles) “move” to the
left as τ increases. The map is an order preserving bijection provided τ1 , τ2 are
in a connected component of (d/2, d) \ Cd , where

Cd = {τc ∈ (d/2, d) : 2τc ∈ Z if d even, 4τc ∈ Z if d odd} (8.6)

However, as τ2 crosses an element of Cd , there is a “flip” in the stratification.

When this flip occurs at δ = 0, this is the phenomenon discovered by Thaddeus
[20]; the discussion here is an extension of this effect to the entire stratification.
 Morse theory and stable pairs 151

Finally, we will also have need to refer to the Harder-Narasimhan stratifica-

tion of the space A of unitary connections on E. We denote this by
,
XjA = Aj  (8.7)
d/2≤j  ≤j

The following statement will be used later on. It is an immediate consequence

of the descriptions of the strata above.
Lemma 8.2.12 Consider the projection pr : B → A. Then
 + 
pr(Bδ ) = Aj (δ) , δ ∈ −
τ,d ∪ τ,d ∩ [τ − d/2, +∞)
A +
pr(Bδ ) = Xd−j (δ) , δ ∈ τ,d ∩ (0, τ − d/2)

8.2.2 Deformation theory

Fix a conformal metric on M, normalized1 for convenience so that
vol(M) = 2π . Infinitesimal deformations of (A, #) ∈ B modulo equivalence
are described by the following elliptic complex, which we denote by C(A,#)
(cf. [4]).

C(A,#)
0 D1
/ C1 D2
/ C2
(A,#) (A,#)

D1
/ 0,1 (End E) ⊕ 0 (E) D2
/ 0,1 (E) (8.8)
0 (End E)
D1 (u) = (−dA u, u#) , D2 (a, ϕ) = dA ϕ + a#
Here, D1 is the linearization of the action of the complex gauge group G C on
B, and D2 is the linearization of the condition dA # = 0. Note that D2 D1 = 0
if (A, #) ∈ B. The hermitian metric gives adjoint operators
D1∗ (a, ϕ) = −(dA )∗ a + ϕ#∗ , D2∗ (β) = (β#∗ , (dA )∗ β) (8.9)
The spaces of harmonic forms are by definition
H0 (C(A,#) ) = ker D1
H1 (C(A,#) ) = ker D1∗ ∩ ker D2
H2 (C(A,#) ) = ker D2∗
Vectors in 0,1 (End E) ⊕ 0 (E) that are orthogonal to the G C -orbit through
(A, #) are in ker D1∗ , and a slice for the action of G C on B is therefore given by
 
S(A,#) = ker D1∗ ∩ (a, ϕ) ∈ 0,1 (End E) ⊕ 0 (E) : D2 (a, ϕ) + aϕ = 0
(8.10)
1 More generally, the scale invariant parameter is τ vol(M)/2π .
152 R. A. Wentworth and G. Wilkin

Define the slice map

 : (ker D1 )⊥ × S(A,#) → B
(8.11)
(u, a, ϕ) → eu · (A + a, # + ϕ)

The proof of the following may be modeled on [21, Proposition 4.12]. We

omit the details.

Proposition 8.2.13 The slice map  is a local homeomorphism from a neigh-

borhood of 0 in (ker D1 )⊥ × S(A,#) to a neighborhood of (A, #) in B.

The Kuranishi map is defined by

k
0,1 (End E) ⊕ 0 (E)−→ 0,1 (End E) ⊕ 0 (E)
k(a, ϕ) = (a, ϕ) + D2∗ ◦ G2 (aϕ)

where G2 denotes the Green’s operator associated to the laplacian D2 (D2 )∗ .

We have the following standard result (cf. [14, Chapter VII] for the case of
holomorphic bundles over a Kähler manifold and [4] for this case).

Proposition 8.2.14 The Kuranishi map k maps S(A,#) to harmonics H1 (C(A,#) ),

and in a neighborhood of zero it is a local homeomorphism onto its image.
Moreover, if H2 (C(A,#) ) = {0}, then k is a local homeomorphism S(A,#) →
H1 (C(A,#) ).

The following is immediate from (8.8) and (8.9).

Lemma 8.2.15 Given (A, #) ∈ B, if # = 0 then H0 (C(A,#) ) = H2 (C(A,#) ) =

{0}. If H 1 (E) = {0} then H2 (C(A,#) ) = {0}.

We will be interested in the deformation complex along higher critical sets of

the Yang-Mills-Higgs functional. As we will see in the next section, in addition
to the Yang-Mills connections (where # ≡ 0), the other critical sets correspond
to split bundles E = L1 ⊕ L2 , (A, #) = (A1 ⊕ A2 , #1 ⊕ {0}), with deg L1 =
j ≥ deg L2 = d − j . Here, j = j (δ) for some δ ∈ + τ,d , or j = d − j (δ) for
some δ ∈ − τ,d . The set of all such critical points will therefore be denoted by
ηδ ⊂ B. We will denote the components of End E  Li ⊗ L∗j in the complex
by uij , aij , ϕij .
In this case, H1 (C(A,#) ) consists of all (a, ϕ) satisfying

(dA )∗ a12 = 0 (d  )∗ a22 = 0

(d  )∗ a11 − ϕ1 #∗1 = 0 (dA )∗ a21 − ϕ2 #∗1 = 0 (8.12)
dA 1 ϕ1 + a11 #1 = 0 dA 2 ϕ2 + a21 #1 = 0
 Morse theory and stable pairs 153

We use this formalism to define deformation retractions in a neighborhood

of (A, #) ∈ B in two cases. First, we have

Lemma 8.2.16 Suppose (A, #) = (A1 ⊕ A2 , #1 ⊕ 0) is a split pair as above,

#1 = 0. Let
neg.
S(A,#) = {(a, ϕ) ∈ S(A,#) : aij = 0 , (ij ) = (21) , and ϕ1 = 0}

S(A,#) = {(a, ϕ) ∈ S(A,#) : (a21 , ϕ2 ) = 0}
neg.
Then there is an equivariant deformation retraction S(A,#) → S(A,#) which
neg. 
restricts to a deformation retraction S(A,#) \ {0} → S(A,#) .

Proof By Lemma 8.2.15 and Proposition 8.2.14, the Kuranishi map gives a
homeomorphism of the slice with H1 (C(A,#) ). Hence, it suffices to define the
retraction there. For this we take

rt (a11 , a12 , a21 , a22 ; ϕ1 , ϕ2 ) = (ta11 , ta12 , a21 , ta22 ; tϕ1 , ϕ2 ) , t ∈ [0, 1]

Notice that this preserves the equations in (8.12). 

Second, near minimal Yang-Mills connections, we find a similar retraction

under the assumption that H2 (C(A,#) ) vanishes.

Lemma 8.2.17 Suppose d > 4g − 4 and A is semistable. Let

neg.
S(A,0) = {(a, ϕ) ∈ S(A,0) : a = 0}

S(A,0) = {(a, ϕ) ∈ S(A,0) : ϕ = 0}
neg.
Then there is an equivariant deformation retraction S(A,0) → S(A,0) which
neg. 
restricts to a deformation retraction S(A,0) \ {0} → S(A,0) .

Proof Let E be the holomorphic bundle given by A. Since E is semistable, so

is E ∗ ⊗ KM , where KM is the canonical bundle of M. On the other hand, by the
assumption, deg(E ∗ ⊗ KM ) = 4g − 4 − deg E < 0. Hence, by Serre duality,
H 1 (E)  H 0 (E ∗ ⊗ KM )∗ = {0}. Given a, let Ha denote harmonic projection

to ker dA+a . It follows that for a in a small neighborhood of the origin in
the slice, Ha is a continuous family. We can therefore define the deformation
retraction explicitly by

rt (a, ϕ) = (ta, Hta (ϕ)) , t ∈ [0, 1]

For a sufficiently small neighborhood of the origin in the slice, this preserves

the set S(A,0) . It is also clearly equivariant. 
154 R. A. Wentworth and G. Wilkin

8.3 Morse theory

8.3.1 The τ -vortex equations
Let μ(A, #) = ∗FA − i##∗ . Then ∗μ is a moment map for the action of G on
B ⊂ A × 0 (E). Let τ > 0 be a positive parameter and define the Yang-Mills-
Higgs functional

fτ (A, #) = μ + iτ · id 2
(8.1)

Solutions to the τ -vortex equations are the absolute minima of fτ :

μ(A, #) + iτ · id = 0 (8.2)

Theorem 8.3.1 (Bradlow [3])

Mτ,d = {(A, #) ∈ B : μ(A, #) + iτ · id = 0} /G.

If the space of solutions to the τ -vortex equations is nonempty, then τ must

satisfy the following restriction.

μ + iτ · id = ∗FA − i##∗ + iτ · id = 0

i (8.3)
=⇒ Tr(∗FA − i##∗ ) = 2τ ⇐⇒ deg E + # 2
= 2τ
2π M
Therefore 2τ ≥ d (with strict inequality if we want to ensure that # = 0).
Theorem 2.1.6 of [3] shows that a solution to the τ -vortex equations which is
not τ -stable must split. Moreover, since rankE = 2 the solutions can only split
if τ is an integer. In particular, for a generic choice of τ solutions to (8.2) must
be τ -stable. In general, critical sets of fτ can be characterized in terms of a
decomposition of the holomorphic structure of E. The critical point equations
for the functional fτ are

dA (μ + iτ · id) = 0 (8.4)

(μ + iτ · id) # = 0 (8.5)

There are three different types of critical points.

(0) Absolute minimum fτ−1 (0).

(I) Yang-Mills connections with # = 0. Then either A is an irreducible
Yang-Mills minimum or E splits holomorphically as E = L1 ⊕ L2 . The
latter exist for all values of deg L1 ≥ d/2 and the existence of the critical
points is independent of the choice of τ . However, as shown below the
Morse index does depend on τ . If E is semistable (resp. deg L1 < τ )
we call this a critical point of type Ia , and we label it δ = τ − d/2
 Morse theory and stable pairs 155

(resp. δ = deg L1 − d + τ ). If deg L1 > τ it is of type Ib , and set δ =

deg L1 − d + τ .
(II) E splits holomorphically as E = L1 ⊕ L2 , and # ∈ H 0 (L1 ) \ {0}. On
L1 we have

∗FA1 − i##∗ = −iτ, # 2

= 2π (τ − deg L1 )

Therefore deg L1 < τ . Further subdivide these depending upon deg L1 .

(II− ) deg L1 ≤ d − τ , δ = d − deg L1 − τ ;
(II+ ) d − τ < deg L1 < τ , δ = deg L1 − d + τ ;

Let S d M denote the d-th symmetric product of the Riemann surface M, and
Jd (M) the Jacobian variety of degree d line bundles on M. For future reference
we record the following

Proposition 8.3.2 For δ ∈ τ,d \ {0},

HG∗ (ηδ ) =
⎧
⎪
⎪ HG∗ (Ass ) Type I, δ = τ − d/2
⎪
⎪
⎨H ∗ (J (M) × J ∗
j (δ) d−j (δ) (M)) ⊗ H (BU (1) × BU (1)) Type I, δ = τ − d/2
⎪H ∗ (S j (δ) M × Jd−j (δ) (M)) ⊗ H ∗ (BU (1))
⎪ Type II+
⎪
⎪
⎩ ∗ d−j (δ)
H (S M × Jj (δ) (M)) ⊗ H ∗ (BU (1)) Type II−

8.3.2 The gradient flow

Consider the negative gradient flow of the Yang-Mills-Higgs functional fτ
defined on the space B ⊂ A × 0 (E). Since the functional is very similar to
that studied in [10], we only sketch the details of the existence and convergence
of the flow and focus on showing that the Morse stratification induced by
the flow is equivalent to the Harder-Narasimhan stratification described in
Section 8.2.1.
The gradient flow equations are
∂A ∂#
= 2 ∗ dA (μ + iτ ) , = −4i(μ + iτ )# (8.6)
∂t ∂t
Theorem 8.3.3 The gradient flow of fτ with initial conditions in B exists for
all time and converges to a critical point of fτ in the smooth topology.

A standard calculation (cf. [3, Section 4]) shows that fτ can be re-written as

 2  2
fτ = |FA |2 + dA # + ##∗  − 2τ |#|2 + |τ |2 dvol + 4τ deg E
X
(8.7)
156 R. A. Wentworth and G. Wilkin

This is very similar to the functional YMH studied in [10], and the proof
for existence of the flow for all positive time follows the same structure (which
is in turn modeled on Donaldson’s proof for the Yang-Mills functional in [9]),
therefore we omit the details. An important part of the proof worth mentioning
here is that the flow is generated by the action of G C , i.e. for all t ∈ [0, ∞)
there exists g(t) ∈ G C such that the solution (A(t), #(t)) to the flow equations
(8.6) with initial condition (A, #) is given by (A(t), #(t)) = g(t) · (A, #).
To show that the gradient flow converges, one can use the results of Theorem
B of [11] (where again, the functional is not exactly the same as fτ , but it has
the same structure and so the proof of convergence is similar). The statement
of [11, Theorem B] only describes smooth convergence along a subsequence
(since they also study the higher dimensional case where bubbling occurs), and
to extend this to show that the limit is unique we use the Lojasiewicz inequality
technique of [19] and [17]. The key estimate is contained in the following
proposition.

Proposition 8.3.4 Let (A∞ , #∞ ) be a critical

 point of fτ . Then there exist
ε1 > 0, a positive constant C, and θ ∈ 0, 12 , such that

(A, #) − (A∞ , #∞ ) < ε1

implies that

∇fτ (A, #) L2 ≥ C |fτ (A, #) − fτ (A∞ , #∞ )|1−θ (8.8)

The proof is similar to that in [21], and so is omitted.

The rest of the proof of convergence then follows the analysis in [21] for
Higgs bundles. The key result is the following proposition, which is the analog
of [21, Proposition 3.7] (see also [19] or [17, Proposition 7.4]).

Proposition 8.3.5 Each critical point (A, #) of fτ has a neighborhood U

such that if (A(t), #(t)) is a solution of the gradient flow equations for fτ
and (A(T ), #(T )) ∈ U for some T , then either fτ (A(t), #(t)) < fτ (A, #)
for some t, or (A(t), #(t)) converges to a critical point (A , # ) such that
fτ (A , # ) = fτ (A, #). Moreover, there exists ε (depending on U ) such that
(A , # ) − (A, #) < ε.

The next step is the main result of this section: The Morse stratification
induced by the gradient flow of fτ is the same as the τ -Harder–Narasimhan
stratification described in Section 8.2.1. First recall the Hitchin–Kobayashi
correspondence from Theorem 8.3.1, and the distance-decreasing result from
[10], which can be re-stated as follows.
 Morse theory and stable pairs 157

Lemma 8.3.6 (Hong [10]) Let (A1 , #1 ) and (A2 , #2 ) be two pairs related by
an element g ∈ G C . Then the distance between the G-orbits of (A1 (t), #1 (t))
and (A2 (t), #2 (t)) is non-increasing along the flow.
Recall that the critical sets associated to each stratum are given in Section
8.3.1, and that the critical set associated to the stratum Bδ is denoted ηδ . Define
Sδ ⊂ B to be the subset of pairs that converge to a point in Cδ under the gradient
flow of fτ . The next lemma gives some standard results about the critical sets
of fτ .

Lemma 8.3.7
(i) The critical set ηδ is the minimum of the functional fτ on the stratum Bδ .
(ii) The closure of each G C orbit in Bδ intersects the critical set ηδ .
(iii) There exists ε > 0 (depending on τ ) such that (A, #) ∈ ηδ and (A , # ) ∈
ηδ with δ = δ  implies that (A, #) − (A , # ) ≥ ε.
Proof Since these results are analogous to standard results for the Yang–Mills
functional (see for example [1], [6], or [7]), and the proof for holomorphic pairs
is similar, we only sketch the idea of the proof here.
r The first statement follows by noting that the convexity of the norm-square
function · 2 shows that the minimum of fτ on each extension class occurs
at a critical point. This can be checked explicitly for each of the types Ia , Ib ,
II+ , and II− .
r To see the second statement, simply scale the extension class and apply
Theorem 8.3.1 (the Hitchin–Kobayashi correspondence) to the graded object
of the filtration (cf. [7, Theorem 3.10] for the Yang–Mills case).
r The third statement can be checked by noting that (modulo the G-action) the
critical sets are compact, and then explicitly computing the distance between
distinct critical sets. 
As a consequence we have

Proposition 8.3.8
(i) Each critical set ηδ has a neighborhood Vδ such that Vδ ∩ Bδ ⊂ Sδ .
(ii) Sδ ∩ Bδ is G C -invariant.
Proof Proposition 8.3.5 implies that there exists a neighborhood Vδ of each
critical set ηδ such that if (A, #) ∈ Vδ then the flow with initial conditions
(A, #) either flows below ηδ , or converges to a critical point close to ηδ . Since
fτ is minimized on each Harder–Narasimhan stratum Bδ by the critical set ηδ ,
the flow is generated by the action of G C , and the strata Bδ are G C -invariant,
then the first alternative cannot occur if (A, #) ∈ Bδ ∩ Vδ . Since the critical
158 R. A. Wentworth and G. Wilkin

sets are a finite distance apart, then (by shrinking Vδ if necessary) the limit
must be contained in ηδ . Therefore Vδ ∩ Bδ ⊂ Sδ , which completes the proof
of (i).
To prove (ii), for each pair (A, #) ∈ Sδ ∩ Bδ , let
 
Y(A,#) = g ∈ G C : g · (A, #) ∈ Sδ ∩ Bδ .

The aim is to show that Y(A,#) = G C . Firstly we note that since the group 
of components of G C is the same as that for the unitary gauge group G, the
flow equations (8.6) are G-equivariant, and the critical sets ηδ are G-invariant,
then it is sufficient to consider the connected component of G C containing the
identity. Therefore the problem reduces to showing that Y(A,#) is open and
closed. Openness follows from the continuity of the group action, the distance-
decreasing result of Lemma 8.3.6, and the result in part (i). Closedness follows
by taking a sequence of points {gk } ⊂ Y(A,#) that converges to some g ∈ G C ,
and observing that the distance-decreasing result of Lemma 8.3.6 implies that
the flow with initial conditions g · (A, #) must converge to a limit close to the
G-orbit of the limit of the flow with initial conditions gk · (A, #) for some large
k. Since the critical sets are G-invariant, and critical sets of different types are a
finite distance apart, then by taking k large enough (so that gk · (A, #) is close
enough to g · (A, #)) we see that the limit of the flow with initial conditions
g · (A, #) must be in ηδ . Therefore Y(A,#) is both open and closed. 

Theorem 8.3.9 The Morse stratification by gradient flow is the same as the
Harder–Narasimhan stratification in Definition 8.2.9.

Proof The goal is to show that Bδ ⊆ Sδ for each δ. Let x ∈ Bδ . By Lemma

8.3.7 (ii) the closure of the orbit G C · x intersects ηδ , therefore there exists
g ∈ G C such that g · x ∈ Vδ ∩ Bδ ⊆ Sδ by Proposition 8.3.8 (i). Since Sδ ∩ Bδ
is G C -invariant by Proposition 8.3.8 (ii), then x ∈ Bδ ∩ Sδ also, and therefore
Bδ ⊆ Sδ . Since {Bδ } and {Sδ } are both stratifications of B, then we have Bδ = Sδ
for all δ. 

Remark 8.3.10 While we have identified the stable strata of the critical sets
with the Harder-Narasimhan strata, the ordering on the set τ,d coming from the
values of YMH is more complicated. Since this will not affect the calculations,
we continue to use the ordering already defined in Section 2.

We may now reformulate the main result, Theorem 8.1.1. The key idea
is to define a substratification of {Xδ , Xδ }δ∈τ,d by combining Bδ and Aj (δ)
for δ ∈ +τ,d ∩ Iτ,d . In other words, this is simply {Xδ }δ∈τ,d . We call this the
modified Morse stratification.
 Morse theory and stable pairs 159

Theorem 8.3.11 The modified Morse stratification {Xδ }δ∈τ,d is G-

equivariantly perfect in the following sense: For all δ ∈ τ,d , the long exact
sequence
· · · −→ HG∗ (Xδ , Xδ1 ) −→ HG∗ (Xδ ) −→ HG∗ (Xδ1 ) −→ · · · (8.9)
splits. Here, δ1 denotes the predecessor of δ in τ,d .

8.3.3 Negative normal spaces

For critical points (A, #) ∈ ηδ , a tangent vector
(a, ϕ) ∈ 0,1 (End E) ⊕ 0 (E)
is an eigenvector for the Hessian of fτ if
i[μ + iτ · id, a] = λa (8.10)
i(μ + iτ · id)ϕ = λϕ (8.11)
neg.
Let V(A,#) ⊂ 0,1 (End E) ⊕ 0 (E) denote the span of all such (a, ϕ) with
λ < 0. This is clearly G-invariant, since fτ is. Let S(A,#) be the slice at (A, #).
neg.
Then we set νδ ∩ S(A,#) = V(A,#) ∩ S(A,#) . Using Proposition 8.2.13, this gives
a well-defined G-invariant subset νδ ⊂ B, which we call the negative normal
space at ηδ . By definition, ηδ is a closed subset of νδ .
We next describe νδ in detail for each of the critical sets:
(Ia ) Recall that in this case # ≡ 0. If E semistable, the negative eigenspace of
the Hessian is H 0 (E). To see this, note that since # = 0 then i(μ + iτ ·
id) = (d/2 − τ ) · id is a negative constant multiple of the identity (by
assumption τ > d/2). Therefore i[μ + iτ · id, a] = 0, and a = 0. Then
the slice equations imply ϕ ∈ H 0 (E). If E = L1 ⊕ L2 , then H2 (C(A,0) )
is nonzero in general. From the slice equations, we see that the negative
eigendirections νδ of the Hessian are given by
dA 2 ϕ2 + a21 ϕ1 = 0 , (a21 , ϕ1 ) ∈ H 0,1 (L∗1 L2 ) ⊕ H 0 (L1 ) (8.12)
(Ib ) This is similar to the case above, except now for negative directions,
ϕ1 ≡ 0. We therefore conclude that νδ is given by
H 0,1 (L∗1 L2 ) ⊕ H 0 (L2 ) (8.13)
Note that if δ > τ , then deg L2 = d − j (δ) < 0, and so νδ−
has constant
∗
dimension dimC H (L1 L2 ) = 2j (δ) − d + g − 1.
0,1

(II+ ) In this case, # ≡ 0, so by Lemma 8.2.15, H2 (C(A,0) ) = 0, and the slice

is homeomorphic to H1 (C(A,0) ) via the Kuranishi map. The negative
160 R. A. Wentworth and G. Wilkin

eigenspace of the Hessian is then just

(dA )∗ a21 − ϕ2 #∗1 = 0 , dA ϕ2 + a21 #1 = 0 (8.14)

(II− ) This is similar to above, except now ϕ2 ≡ 0. Hence, the fiber of νδ is

given by

H 0,1 (L∗2 L1 ) (8.15)

Note that dimC H 0,1 (L∗2 L1 ) = 2j (δ) − d + g − 1.

To see (II+ ) and (II− ), we need to compute the solutions to (8.10) and (8.11),
which involves knowing the value of i(μ + iτ · id) on the critical set. Equation
(8.4) shows that
λ1 0
i(μ + iτ · id) =
0 λ2

where λ1 ∈ 0 (L∗1 L1 ) and λ2 ∈ 0 (L∗2 L2 ) are constant. Since # ∈ H 0 (L1 ) \

{0}, then (8.5) shows that λ1 = 0. Since λ2 is constant, the integral over M
becomes
  
1 i 1
λ2 = λ2 dvol = FA − τ dvol = deg L2 − τ
2π M 2π M 2 2π M
Therefore, if d − τ < deg L1 = d − deg L2 , then deg L2 < τ and so λ2 is nega-
tive. Similarly, if deg L1 < d − τ then λ2 is positive. Equation (8.10) then shows
that a ∈ 0,1 (L∗1 L2 ) if d − τ < deg L1 , and a ∈ 0,1 (L∗2 L1 ) if deg L1 < d − τ .
Similarly, if d − τ < deg L1 then ϕ ∈ 0 (L2 ), and if deg L1 < d − τ then
ϕ = 0. Equations (8.14) and (8.15) then follow from the slice equations.
The following lemma describes the space of solutions to (8.12) when ϕ1 is
fixed.

Lemma 8.3.12 Fix ϕ1 . When ϕ1 = 0 then the space of solutions {(a21 , ϕ2 )} to

(8.12) is isomorphic to H 0,1 (L∗1 L2 ) ⊕ H 0 (L2 ). When ϕ1 = 0 then the space of
solutions {(a21 , ϕ2 )} to (8.12) has dimension deg L1 .

Proof The first case (when ϕ1 = 0) is easy, since the equations for a ∈
0,1 (L∗1 L2 ) and ϕ2 ∈ 0 (L2 ) become

dA∗ a = 0, dA ϕ2 = 0. (8.16)

In the second case (when ϕ1 = 0 is fixed), note (8.12) implies that H(aϕ1 ) =
0, where H denotes the harmonic projection 0,1 (L2 ) → H 0,1 (L2 ). Hence, it
suffices to show that the map

H 0,1 (L∗1 L2 ) → H 0,1 (L2 ) (8.17)

 Morse theory and stable pairs 161

given by multiplication with ϕ1 (followed by harmonic projection) is surjective.

For then, since deg L∗1 L2 < 0, we have by Riemann–Roch that the dimension
of (8.12) is h0 (L2 ) + h1 (L∗1 L2 ) − h1 (L2 ) = deg L1 . By Serre duality, (8.17)
is surjective if and only if H 0 (KL∗2 ) → H 0 (KL∗2 L1 ) is injective. But since
ϕ1 = 0, multiplication gives an injection of sheaves O → L1 , and the result
follows by tensoring and taking cohomology. 
Lemma 8.3.13 The space of solutions to (8.14) has constant dimension =
deg L1 = j (δ).
Proof Consider the subcomplex C(A,#)
LT

0 (L∗1 L2 ))
D1
/ 0,1 (L∗ L2 ) ⊕ 0 (L2 ) D2
/ 1 (L2 ) (8.18)
1

Since # = 0, by Lemma 8.2.15 the cohomology at the ends of the complex

(8.18) vanishes, and we have (by Riemann–Roch)
dimC H1 (C(A,#)
LT
) = dimC (ker D1∗ ∩ ker D2 )
= h1 (L∗1 L2 ) + h0 (L2 ) − h1 (L2 ) − h0 (L∗1 L2 )
= − deg L∗1 L2 + g − 1 + deg L2 + (1 − g)
= deg L1 
We summarize the the above considerations with
Corollary 8.3.14 The fiber of νδ is linear of constant dimension for critical
sets of type II± , and for those of type Ib provided δ ∈ + τ,d ∩ [τ − d/2, τ ]. The
complex dimension of the fiber in these cases is σ (δ), where
6
2j (δ) − d + g − 1 if type Ib or II−
σ (δ) =
j (δ) if type II+

Remark 8.3.15 The strata for δ ∈ Iτ,d have two components corresponding
to the strata Aj (δ) and Bδ . When there is a possible ambiguity, we will dis-
tinguish these by the notation νI,δ for the negative normal spaces to strata of
type Ia or Ib , and νI I,δ for the negative normal spaces to strata of type II+
or II− .

8.3.4 Cohomology of the negative normal spaces

As in [8], at certain critical sets – namely, those of type Ia , Ib where δ ∈ +
τ,d ∩
[τ − d/2, τ ] – the negative normal directions are not necessarily constant in
162 R. A. Wentworth and G. Wilkin

dimension. In the present case, they are not even linear. In order to carry out
the computations, we appeal to a relative sequence by considering special
subspaces with better behavior.

Definition 8.3.16 For δ ∈ + τ,d ∩ (τ − d/2, τ ], let νI,δ be the negative normal
space to a critical set with # ≡ 0, as in Section 8.3.3. Define


 
νI,δ = (a, ϕ1 , ϕ2 ) ∈ νI,δ : (a, ϕ1 , ϕ2 ) = 0

 
νI,δ = (a, ϕ1 , ϕ2 ) ∈ νI,δ : a = 0

The goal of this section is the proof of the following

Proposition 8.3.17

∗−2(2j (δ)−d+g−1)
δ ∈ + ∗ 
τ,d ∩ (τ − d/2, τ ] : HG (νI,δ , νI,δ )  HS 1 ×S 1 (ηjA(δ) )
(8.19)
δ∈ +
τ,d ∩ (2τ − d, τ ] : HG∗ (νI,δ
 
, νI,δ )
∗−2(2j (δ)−d+g−1) d−j (δ)
 HS 1 (S M × Jj (δ) (M))
(8.20)
∗−2j (δ)
δ ∈ + ∗  
τ,d ∩ (τ −d/2, 2τ −d) : HG (νI,δ , νI,δ )  HS 1 (S j (δ) M ×Jd−j (δ) (M))
(8.21)
∗−2(2j (δ)−d+g−1)
⊕ HS 1 (S d−j (δ) M × Jj (δ) (M))

Proof Fix E = L1 ⊕ L2 . Consider first the case τ > deg L1 = j (δ) > d/2,
and deg L2 = d − j (δ) < d/2. Define the following spaces

 
ωδ = (A1 , A2 , a, ϕ1 , ϕ2 ) ∈ νI,δ : (a, ϕ2 ) = 0
 
Zδ− = (A1 , A2 , a, ϕ1 , ϕ2 ) ∈ νI,δ : ϕ1 = 0
 
Zδ = (A1 , A2 , a, ϕ1 , ϕ2 ) ∈ νI,δ : ϕ1 = 0, (a, ϕ2 ) = 0
 
Yδ = (A1 , A2 , a, ϕ1 , ϕ2 ) ∈ νI,δ : ϕ1 = 0
 
Yδ = (A1 , A2 , a, ϕ1 , ϕ2 ) ∈ νI,δ : ϕ1 = 0, (a, ϕ2 ) = 0
 
Tδ = (A1 , A2 , a, ϕ1 , ϕ2 ) ∈ νI,δ : ϕ1 = 0, (a, ϕ2 ) = 0
 Morse theory and stable pairs 163

Note that Yδ = νI,δ \ Zδ− = νI,δ


\ Zδ and Yδ = ωδ \ Zδ . Consider the following
commutative diagram.

..
.


··· / H p (νI,δ , ν  ) / H p (νI,δ ) / H p (ν  ) / ···
G G I,δ
I,δ
o 7 G
ξ oooo
ooo
 ooo
p 
HG (νI,δ , νI,δ )
OOO
OOOξ 
OOO
 OO'
... / H p (ν  , ωδ ) / H p (ν  , ν  ) β / H p (ωδ , ν  ) / ···
G I,δ G I,δ I,δ G I,δ


..
.
(8.22)
r First, it follows as in the proof of [8, Thm. 2.3] that the pair (νI,δ , νI,δ

) is
A A
homotopic to the Atiyah–Bott pair (Xj (δ) , Xj (δ)−1 ). Hence, (8.19) follows
from [1].
r Consider the pair (νI,δ

, ωδ ). Excision of Zδ gives the isomorphism

HG∗ (νI,δ

, ωδ ) ∼
= HG∗ (νI,δ

\ Zδ , ωδ \ Zδ ) ∼
= HG∗ (Yδ , Yδ ) (8.23)

The space Yδ = Yδ \ Tδ , and Lemma 8.3.12 shows that Yδ is a vector bundle
over Tδ with fibre dimension = deg L1 . Therefore the Thom isomorphism
implies

HG∗ (Yδ , Yδ ) = HG∗ (Yδ , Yδ \ Xδ ) ∼

∗−2j (δ)
= HG (Tδ )

and therefore
∗−2j (δ)
HG∗ (νI,δ

, ωδ ) = HG∗ (Yδ , Yδ ) = HS 1 (S j (δ) M × Jd−j (δ) (M)) (8.24)
r Consider (ωδ , νI,δ

). By retraction, the pair is homotopic to the intersection
with ϕ1 = 0. It then follows exactly as in [8] (or the argument above) that

)∼
∗−2(2j (δ)−d+g−1) d−j (δ)
HG∗ (ωδ , νI,δ

= HS 1 (S M × Jj (δ) (M)) (8.25)

(Recall that dim H 0,1 (L∗1 L2 ) = 2j (δ) − d + g − 1 by Riemann–Roch, and

that deg L2 = d − j (δ)).
164 R. A. Wentworth and G. Wilkin

It then follows as in [8] that ξ  , and hence also β, is surjective. This implies
that the lower horizontal exact sequence splits, and (8.21) follows from (8.24)
and (8.25). This completes the proof in this case. The case where deg L1 > τ

is simpler, since ϕ1 ≡ 0 from (8.11). Hence, ωδ = νI,δ , and the proof proceeds
as above. 

8.3.5 The Morse-Bott lemma

In this section we prove the fundamental relationship between the relative
cohomology of successive strata and the relative cohomology of the negative
normal spaces. From this we derive the proof of the main result. In the following
we use δ1 to denote the predecessor of δ in τ,d .
 
Theorem 8.3.18 For all δ ∈ τ,d \ + τ,d ∩ Iτ,d ,

HG∗ (Xδ , Xδ1 )  HG∗ (νδ , νδ ) (8.26)

For all δ ∈ +
τ,d ∩ Iτ,d ,

HG∗ (Xδ , Xδ1 )  HG∗ (νI I,δ , νI I,δ ) (8.27)

For all δ ∈ +
τ,d ∩ Iτ,d , δ = τ − d/2,

HG∗ (Xδ , Xδ )  HG∗ (νI,δ , νI,δ


) (8.28)

Eq. (8.28) also holds for δ = τ − d/2, provided d > 4g − 4. In the statements
above, δ1 denotes the predecessor of δ in τ,d .

First, we give a proof of (8.26) in the case δ ∈ + τ,d ∩ [τ − d/2, τ ]. By

excision and convergence of the gradient flow, there is a neighborhood U of ηδ
such that
r U is G-invariant;
r U is the union of images of slices S(A,#) , where (A, #) ∈ ηδ ;
r HG∗ (Xδ , Xδ1 )  HG∗ (U, U \ (U ∩ Bδ ))

Notice that for each slice S(A,#) ∩ U \ (U ∩ Bδ ) = S(A,#) ∩ U , where the lat-
ter is defined as in Lemma 8.2.16. By the lemma, it follows that the pair
(U, U \ (U ∩ Bδ )) locally retracts to (νδ , νδ ). On the other hand, by Corollary
8.3.14, νδ is a bundle over ηδ . It follows by continuity as in [2], that there is
a G-equivariant retraction of the pair (νδ , νδ ) → (U, U \ (U ∩ Bδ )). The result
therefore follows in this case. We also note that by Corollary 8.3.14 and the
Thom isomorphism,

HG∗ (νδ , νδ )  HG∗−2σ (δ) (ηδ ) (8.29)

 Morse theory and stable pairs 165

Remark 8.3.19 Notice that by Corollary 8.3.14 the same argument also proves
(8.27). For d > 4g − 4, we can use Lemma 8.2.17 in the same way to derive
(8.28) for δ = τ − d/2. In this case, by the Thom isomorphism, we have
∗−2(d+2−2g)
HG∗ (Xτ −d/2 , Xτ −d/2 )  HG (Ass ) (8.30)

Lemma 8.3.20 For δ ∈ + τ,d ∩ [τ − d/2, τ ], or if δ = τ − d/2 and d > 4g −

4, then the long exact sequence (8.9) splits. Similarly, the long exact sequence

· · · −→ HG (Xδ , Xδ1 ) −→ HG (Xδ ) −→ HG (Xδ1 ) −→ · · ·

p p p
(8.31)

splits for all δ ∈ Iτ,d .

Proof Indeed, since (8.26) holds in this case, we have

··· / H p (Xδ , Xδ1 ) α / H p (Xδ ) / H p (Xδ1 ) / ··· (8.32)

G G G

∼
=
 
HG (νδ , νδ )
p β
/ H p (ηδ )
G

Now νδ → ηδ is a complex vector bundle with a G-action and a circle subgroup

that fixes ηδ and acts freely on νδ \ ηδ , so by [1, Prop. 13.4], β is injective. It
follows that is α is injective as well, and hence the sequence splits. The second
statement follows by Remark 8.3.19 and the same argument as above. 

It remains to prove (8.28) and the remaining cases of (8.26). As noted

above, in these cases the negative normal spaces are no longer constant in
dimension, and indeed they are not even linear in the fibers. From the point of
view of deformation theory, the Kuranishi map near these critical sets is not
surjective, and defining an appropriate retraction is more difficult than in the
situation just considered. Instead, we resort to the analog of the decomposition
used in Section 8.3.3. Let Xδ = Xδ \ pr−1 (Aj (δ) ). Note that by Lemma 8.2.12,
Xδ ⊂ Xδ . We will prove the following

Proposition 8.3.21 Suppose δ ∈ +

τ,d ∩ (τ − d/2, τ ]. Then

HG∗ (Xδ , Xδ ) ∼

= HG∗ (νI,δ , νI,δ

)) (8.33)
∗   ∼ ∗  
H (X , X ) = H (ν , ν )
G δ δ G I,δ I,δ (8.34)

Proof of (8.33) By [1] and (8.19), it suffices to prove

HG∗ (Xδ , Xδ ) ∼

= HG∗ (XjA(δ) , XjA(δ)−1 ) (8.35)

We first note that the pair (Xδ , Xδ ) is not necessarily invariant under scaling t#,
t → 0, in particular because of the strata in − τ,d (cf. Lemma 8.2.12). However,
166 R. A. Wentworth and G. Wilkin

if we set
,
3δ = Xδ ∪
X 3δ = X
XδA +τ , X 3δ \ pr−1 (Aj (δ) )
δ  ≤δ, δ  ∈−
τ,d

then by excision on the closed subset

,
Aδ +τ
j (δ)−τ <δ  ≤δ
δ  ∈−
τ,d

3δ , X
it follows that HG∗ (Xδ , Xδ ) = HG∗ (X 3δ ). Then for the pair (X
3δ , X
3δ ), pro-
jection to A is a deformation retraction (by scaling the section #), and we
have
3δ ), pr(X
HG∗ (Xδ , Xδ ) = HG∗ (pr(X 3δ )) (8.36)

Next, let
 , ,  ,
Kδ = pr X3(τ −d/2) ∪ Bδ ∪ Bδ ∩ Ak
δ  ≤δ , δ  ∈−
τ,d δ  <τ −d/2 , δ  ∈+
τ,d
k>j (δ)

Note that Kδ ⊂ pr(X 3δ ). We claim that it is actually a closed subset of pr(X
3δ ).
Indeed, suppose (Ai , #i ) ∈ X(τ −d/2) , (Ai , #i ) → (A, #) ∈ X 3δ , and suppose
that μ+ (Ai ) > j (δ) for each i. By semicontinuity, it follows that μ+ (A) >
j (δ). On the other had, either A ∈ Kδ or (A, #) ∈ Bδ , τ − d/2 < δ  ≤ δ and
δ  ∈ +τ,d . But by Lemma 8.2.12, this would imply A ∈ Aj (δ  ) ; which is a
contradiction, since j (δ  ) ≤ j (δ). It follows that the latter cannot occur, and
hence, Kδ is closed. Similarly,
 , , 
3δ ) = pr X
pr(X 3(τ −d/2) ∪ Bδ ∪ Bδ
δ  ≤δ , δ  ∈−
τ,d δ  <τ −d/2 , δ  ∈+
τ,d
, ,
∪ Ak ∪ pr(Bδ )
d/2<k≤j (δ) τ −d/2<δ  ≤δ , δ  ∈+
τ,d
,
= Kδ ∪ Ak
d/2≤k≤j (δ)

and the union is disjoint. It follows also that

,
pr(X 3δ ) = Kδ ∪ Ak
d/2≤k<j (δ)

Hence, pr(X3δ ) \ Kδ = XjA(δ) , pr(X

3j(δ) ) \ Kδ = XjA(δ)−1 , and (8.35) follows from
(8.36) by excision. 
 Morse theory and stable pairs 167

Proof of (8.34) First consider the case δ ∈ +

τ,d ∩ (τ − d/2, 2τ − d). We have
 , , 
Xδ = X(τ −d/2) ∪ Bδ ∪ Bδ \ pr−1 (Aj (δ) )
δ  ≤δ , δ  ∈−
τ,d δ  <τ −d/2 , δ  ∈+
τ,d
, ,
∪ Ak ∪ Bδ
d/2<k<j (δ) τ −d/2<δ  <δ , δ  ∈+
τ,d

whereas Xδ = Xδ1 ∪ Bδ , where δ1 is the predecessor of δ in τ,d . Also, Xδ =

Xδ1 \ pr−1 (Aj (δ) ). We then have the following diagram

··· / H p (X , X ) / H p (X ) / H p (X ) / ··· (8.37)

G δ δ G δ G δ

f g ∼
=
  
··· / H p (Xδ1 , X ) / H p (Xδ1 ) / H p (X ) / ···
G δ G G δ

where f and g are induced by the inclusion Xδ1 → Xδ . By Lemma 8.3.20 and
(8.27) (see Remark 8.3.19), it follows that g is surjective and
∗−2j (δ) ∗−2j (δ)
ker g = HG∗ (νI I,δ , νI I,δ )  HG (Bδ )  HS 1 (S j (δ) M × Jd−j (δ) M)
by Thom isomorphism. Chasing through the diagram, it follows that f is also
surjective with the same kernel. We conclude that
∗−2j (δ)
HG∗ (Xδ , Xδ )  HG∗ (Xδ1 , Xδ ) ⊕ HS 1 (S j (δ) M × Jd−j (δ) M) (8.38)
It remains to compute the first factor on the right hand side. To begin, notice
that
, , ,
Ak ∪ Bδ ∪ Bδ
d/2<k<j (δ) τ −d/2<δ  <δ , δ  ∈−
τ,d τ −d/2<δ  <δ , δ  ∈+
τ,d

is contained in Xδ and closed in Xδ1 . It follows by excision that

HG∗ (Xδ1 , Xδ )  HG∗ (Xτ −d/2 , Xτ −d/2 \ pr−1 (Aj (δ) ))
Next, we observe that
,
Ass ∪ Bδ
δ  <τ −d/2 , δ  ∈−
τ,d

is contained in Xτ −d/2 \ pr−1 (Aj (δ) ) and closed in Xτ −d/2 . This is clear for Ass .
More generally, if (E, #) in this set and # ≡ 0, then μ+ (E) > τ > j (δ), and
elements in the strata of type II− cannot specialize to points in II+ . Again
applying excision, we have
HG∗ (Xδ1 , Xδ )  HG∗ (Yδ , Yδ \ pr−1 (Aj (δ) ))
168 R. A. Wentworth and G. Wilkin

where
,
Yδ = Bss
τ
∪ Bδ
0<δ  ≤τ −d/2 , δ  ∈+
τ,d

We make a third excision of the closed set

,
Bδ
τ −j (δ)<δ  ≤τ −d/2 , δ  ∈+
τ,d

and a final excision of the subset

 τ ,   , −1 
Dδ = Bss ∪ Bδ ∩ pr (Ak )
0<δ  ≤τ −j (δ) , δ  ∈+
τ,d
k>j (δ)

Notice that
 τ , 
Bss ∪ Bδ \ Dδ = Bss
j (δ)

0<δ  ≤τ −j (δ) , δ  ∈+

τ,d

We conclude that
HG∗ (Xδ1 , Xδ )  HG∗ (Bss
j (δ)
, Bss
j (δ)
\ pr−1 (Aj (δ) ))
Choose ε > 0 small, and let τ  = j (δ) − ε. Then with respect to the τ  -
τ 
τ
stratification, the right hand side above is  HG∗ (Bss ∪ Bετ , Bss ) where ε ∈ −
τ
is the lowest τ  -critical set. Since ε < τ  − d/2, it follows from Lemma 8.3.20
that the long exact sequence (8.9) splits for this stratum. Hence, we have
   ∗−2(2j (δ)−d+g−1)
HG∗ (Xδ1 , Xδ )  HG∗ (Bss
τ
∪ Bετ , Bss
τ
)  HS 1 (S d−j (δ) M ×Jj (δ) (M))
(8.39)
(notice that jτ  (ε) = jτ (δ)). Eqs. (8.38) and (8.39), combined with Propo-
sition 8.3.17, complete the proof. In case δ ∈ Iτ,d , note that by definition
HG∗ (Xδ , Xδ )  HG∗ (Xδ1 , Xδ ). The part of the proof following (8.38) now applies
verbatim to this case. 
Proof of Theorem 8.3.18 For δ ∈ + τ,d ∩ [τ − d/2, τ ], or δ = τ − d/2 and
d > 4g − 4, we have proven the result directly (see the discussion following
Theorem 8.3.18 and also Remark 8.3.19). For δ ∈ + τ,d ∩ (τ − d/2, τ ], the
result follows from Proposition 8.3.21 and the five lemma. 

8.3.6 Perfection of the stratification for large degree

Note that Lemma 8.3.20 shows that the long exact sequence (8.9) splits for all
δ∈/ +τ,d ∩ [τ − d/2, τ ], and also for δ = τ − d/2 if d > 4g − 4. Therefore it
remains to show that (8.9) splits for δ ∈ +τ,d ∩ (τ − d/2, τ ].
 Morse theory and stable pairs 169

Figure 8.2

Firstly we consider the case where δ ∈ + τ,d ∩ [2τ − d, τ ], which corre-

sponds to a stratum of type Ib . Proposition 8.3.21 shows that the vertical long
exact sequence splits and the map ξ is injective in the following commutative
diagram (Figure 8.2).
Therefore the map α p is injective, and so the horizontal long exact sequence
splits also.
Next, suppose δ ∈ + τ,d ∩ (τ − d/2, 2τ − d). For this we need the following
lemma.

Lemma 8.3.22 When δ ∈ + τ,d ∩ (τ − d/2, 2τ − d), then the isomorphisms

HG∗ (Xδ , Xδ ) ∼= HG∗ (νI,δ , νI,δ

) and HG∗ (Xδ1 , Xδ ) ∼
= HG∗ (ωδ , νI,δ

) in equivari-

ant cohomology are induced by an inclusion of triples (νI,δ , ωδ , νI,δ ) →

(Xδ , Xδ1 , Xδ ).

Proof The first isomorphism is contained in (8.33). To see the second iso-
morphism, note that the results of the last section show that HG∗ (Xδ1 , Xδ ) ∼ =
τ 
τ
HG∗ (Bss ∪ Bετ , Bss ), where ε ∈ −
τ  is the lowest τ 
critical set. Excise all but

a neighborhood of Bετ , and deformation retract # so that # is small. Call
these new sets W and W0 , respectively. Then
τ 
τ ∼
HG∗ (Bss ∪ Bετ , Bss ) = HG∗ (W, W0 )

Since # = 0, then we can apply Lemma 8.2.16 to the slices within the

spaces W and W0 , and the resulting spaces are homeomorphic to ωδ and νI,δ
respectively. 
170 R. A. Wentworth and G. Wilkin

Figure 8.3

−
The previous lemma together with the surjection ξ  : HG∗ (νI,δ 
, νI,δ )→
HG∗ (ωδ , νI,δ

) 
from (8.22) implies that the map ξg is surjective in the following
commutative diagram (Figure 8.3).
The isomorphism (8.35) together with the results of [1] show that the map
ξg is injective, and so the same argument as before shows that the horizontal
long exact sequence splits.

8.3.7 The case of low degree

By the results of the previous section, there is only one critical stratum
unaccounted for on the way to completing the proof of Theorem 8.3.11 for
1 ≤ d ≤ 4g − 4. Namely, we need to analyze what happens when we attach the
minimal Yang-Mills stratum Ass , which is the lowest critical set of Type I. More
precisely, from (8.4), we need to show that the inclusion Xτ −d/2 → Xτ −d/2
induces a surjection in G-equivariant rational cohomology for all τ ∈ (d/2, d).
Notice that by (8.5), Xτ −d/2 = Xδ1 for d odd, so this is precisely what we need
to prove; and if d is even, then the above statement together with Lemma 8.3.20
will prove that Xδ1 → Xτ −d/2 induces a surjection in G-equivariant rational
cohomology in this case as well.
In low degree, the negative normal directions exist only over a Brill-Noether
subset of Ass , whose cohomology is unknown, and the dimension of the fiber
jumps in a complicated way; it is not even clear that there is a good Morse-Bott
lemma of the type (8.28) in this case.
 Morse theory and stable pairs 171

Hence, in order to prove surjectivity in this case we will use an indirect

argument via embeddings of the space of pairs of degree d into corresponding
pairs of larger degree. More precisely, this is defined as follows. Choose a point
p ∈ M, and let O(p) denote the holomorphic line bundle with divisor p. We
also choose a hermitian metric on O(p). Choose a holomorphic section σp of
O(p) with a simple zero at p. Note that σp is unique up to a nonzero multiple.
A holomorphic (and hermitian) structure on the complex vector bundle E
induces one on the bundle E = E ⊗ O(p). Moreover, if # ∈ H 0 (E), then # =
0 
# ⊗ σp ∈ H (E). The unitary gauge group G of E is canonically isomorphic
 Hence, we have a G-equivariant embedding B(E) → B(E).
to that of E.  For
simplicity, we will use the notation B = B(E) and B  = B(E).

Let d̃ = d + 2 and τ̃ = τ + 1. Then we note the following properties:
 = d̃
deg E τ̃ ,d̃ = τ,d

deg # = deg # + 1 Iτ̃ ,d̃ = Iτ,d
 = μ+ (E) + 1 jτ̃ ,d̃ (δ) = jτ,d + 1
μ+ (E)

It follows easily that the inclusion respects the Harder-Narasimhan stratifica-

tion, i.e. for all δ ∈ τ,d , Bδ → B δ , Xδ → Xδ , and Xδ → Xδ , where the tilde’s
have the obvious meaning. In particular, if we fix τmax = d − ε, for ε small,
then Bss τmax
→B ss
τ̃max
. Notice that while Bss τmax
gives the “last” moduli space in
the sense that there are no critical values between τmax and d (provided ε is
sufficiently small), B ss
τ̃max
gives the “second to last” moduli space in the sense
that there is precisely one critical value between τ̃max and d̃.

Lemma 8.3.23 The inclusion Bss τmax ss

→B τ̃max
induces a surjection in G-
equivariant rational cohomology.

Proof Since τ is generic, it suffices to prove the result on the level of moduli
spaces, i.e. that the inclusion ı : Mτmax ,d → M τ̃ ,d̃ induces a surjection in
max
cohomology. Consider the determinant map (E, #) → det E. We have the
following diagram

Mτmax ,d
ı /M
 (8.40)
τ̃max ,d̃

det det
 
Jd (M)
j
/ Jd̃ (M)

Now Mτmax ,d is the projectivization of a vector bundle (cf. [20]). Hence, by

the Leray–Hirsch theorem its cohomology ring is generated by the embedding
(det)∗ (H ∗ (Jd (M))), and a 2-dimensional class generating the cohomology of
the fiber. Since ı ∗ (det)∗ = (det)∗ j ∗ , and j ∗ is an isomorphism, it follows that ı ∗
172 R. A. Wentworth and G. Wilkin

is surjective onto (det)∗ (H ∗ (Jd (M))). It remains to show that the 2-dimensional
class is in the image of ı ∗ . But since ı is holomorphic and Mτ̃max ,d̃ is projective,
the Kähler class of Mτ̃max ,d̃ restricted to the image generates the cohomology
of the fiber. 

Lemma 8.3.24 Suppose δ ∈ τmax ,d , δ < τmax − d/2. Then the inclusion
Xδ → Xδ induces a surjection in G-equivariant rational cohomology. The

same holds for Xτ −d/2 → X . τ̃ −d̃/2

Proof By Lemma 8.3.23, the result holds for the semistable stratum. Fix
δ < τ − d/2, and let δ1 be its predecessor in τmax ,d . By induction, we may
assume the result holds for δ1 . By Lemma 8.3.20 we have the following diagram:

0 / H p (X
δ , X
δ1 ) / H p (X
δ ) / H p (X
δ1 ) /0 (8.41)
G G G

f g h
  
0 / H p (Xδ , Xδ1 ) / H p (Xδ ) / H p (Xδ1 ) /0
G G G

By the inductive hypothesis, h is surjective. On the other hand, by (8.26) and

(8.29), surjectivity of f is equivalent to surjectivity of the map HG∗ (η̃δ ) →
HG∗ (ηδ ). From the description of critical sets (cf. Proposition 8.3.2), this map
is induced by the inclusion S j (δ) M → S j (δ)+1 M. Surjectivity then follows by
the argument in [8, Sect. 4]. Since both f and h are surjective, so is g. The
result for any δ < τ − d/2 now follows by induction. If d is even, the exact
same argument, with δ1 = the predecessor of τ − d/2, proves the statement for
Xτ −d/2 as well. 

Lemma 8.3.25 Suppose the inclusion X  τ̃ −d̃/2 induces a sur-

→X
τ̃max −d̃/2 max
jection in G-equivariant rational cohomology. Then the same is true for the
inclusion Xτ max −d/2 → Xτmax −d/2 .

Proof Consider the diagram

τ̃ −d̃/2 )
p
HG (X 
/ HGp (X ) /0 (8.42)
max τ̃max −d̃/2

h
 
p
HG (Xτmax −d/2 ) / HGp (Xτ ) / ···
max −d/2

By Lemma 8.3.24, h is surjective. The result then follows immediately. 

Lemma 8.3.26 Suppose the inclusion Xτ −d/2 → Xτ −d/2 induces a surjection
in G-equivariant rational cohomology for τ = τmax . Then the same is true for
 Morse theory and stable pairs 173

all τ ∈ (d/2, d). Moreover, dim HG (Xτ −d/2 , Xτ −d/2 ) is independent of τ for
p

all p.

Proof The sets Xτ −d/2 , Xτ −d/2 remain unchanged for τ in a connected compo-
nent of (d/2, d) \ Cd , where Cd is given in (8.6). Fix τc ∈ Cd , 2τc − d/2 = k ∈
Z, and let τl < τc < τr be in components (d/2, d) \ Cd containing τc in their
closures. Let δ l,r = 2τc − d/2 − τl,r . Note that δ l,r ∈ −τl,r ,d , δ > τl − d/2,
l

and δ < τr − d/2. Also, we claim

r

Xτr −d/2 = Xτl −d/2 ∪ Bδτll , Xτ r −d/2 = Xτ l −d/2 ∪ Bδτll (8.43)

To see this, we refer to Figure 8.1 and the discussion preceding it. Under
the map τl ,d → τr ,d , δ l → δ r and τl − d/2 → τr − d/2. The claim then
follows if we show that δ r is the predecessor of τr − d/2 in τr ,d , and δ l is
the successor of τl − d/2 in τl ,d (see Figure 8.1). So suppose δ ∈ τr ,d , δ <
τr − d/2. By Remark 8.2.7, we may assume δ ∈ − τr ,d . Write δ + τr =  ∈ Z.
Then  ≤ 2τr − d/2, which implies  ≤ k, and δ ≤ δr . The reasoning is similar
for δl .
Now since the result holds by assumption for τmax , we may assume by
induction that the result holds for τ ≥ τr . Then we have

0 / HGp (Xτr −d/2 , Xτ −d/2 ) / H p (Xτr −d/2 ) / HGp (Xτ −d/2 ) /0
r G r

f g h
  
··· / HGp (Xτl −d/2 , Xτ −d/2 ) / H p (Xτl −d/2 ) / HGp (Xτ −d/2 ) / ···
l G l

(8.44)
By (8.43) and the proof of Lemma 8.3.24, h is surjective. Hence, the lower long
exact sequence must split. Moreover, g is surjective as well, and ker g = ker h.
As a consequence, f must be an isomorphism. The result now follows by
induction. 

Proof of Theorems 8.3.11 and 8.1.2 We proceed by induction as follows.

First, if d > 4g − 4, then by Lemma 8.3.20, the hypothesis of Lemma 8.3.25
is satisfied. It then follows from Lemma 8.3.26 that the inclusion Xτ −d/2 →
Xτ −d/2 induces a surjection in G-equivariant rational cohomology for any τ . In
particular, this is true for the value τ̃max corresponding to degree d − 2. Hence,
the inductive hypothesis holds, and the result is proven for all d. Kirwan
surjectivity follows immediately. 

Proof of Theorem 8.1.3 This follows from Kirwan surjectivity, but more gen-
erally we prove this on each stratum. Clearly it suffices to prove the result for
174 R. A. Wentworth and G. Wilkin

k = 1. Since the gauge groups for E and E  are canonically isomorphic, it suf-
fices by induction to show that if the result holds for the inclusion Xδ → X δ ,

then it also holds for Xδ1 → Xδ1 , where δ1 is the predecessor of δ in τ,d .
By Theorem 8.3.11, the diagram (8.41) holds for all δ. It follows that if g is
surjective, then so is h. This completes the proof. 

8.4 Cohomology of moduli spaces

8.4.1 Equivariant cohomology of τ -semistable pairs
The purpose of this section is to complete the calculation of the G-equivariant
Poincaré polynomial of Bss τ
. First we consider the case where τ is generic.
Choose an integer N , d/2 < N ≤ d, and let τ ∈ (max{d/2, N − 1}, N). Then
the different allowable values of δ for each type of stratum and the cohomology
are as follows (see Proposition 8.3.2).
d/2
(Ia ) There is one stratum Ia corresponding to Ass (indexed by j = d/2),
d/2
and by Lemma 8.3.26 the contribution Ia (t) to the Poincaré polynomial
is independent of τ . For d > 4g − 4 it follows from (8.30) that
t 2d+4−4g G
a (t) =
Id/2 P (Ass ) (8.1)
(1 − t 2 ) t
d/2
where G is defined in [1, p. 577]. We compute Ia (t) in general in Lemma
8.4.3 below. The remaining strata are indexed by integers j = j (δ) = μ+
such that d/2 < j ≤ N − 1 and δ = j − d + τ . The contribution to the
G-equivariant Poincaré polynomial is
t 2(2j (δ)−d+g−1)  
Ija (t) = Pt Jj (δ) (M) × Jd−j (δ) (M)
(1 − t ) 2 2

t 2j (δ)  
− Pt S j (δ) M × Jd−j (δ) (M) (8.2)
1−t 2

t 2(2j (δ)−d+g−1)  d−j (δ) 

− Pt S M × Jj (δ) (M)
1 − t2
(Ib ) There are an infinite number of strata indexed by integers j = j (δ) = μ+
such that N ≤ j and δ = j − d + τ . The contribution is

j t 2(2j (δ)−d+g−1)  
Ib (t) = Pt Jj (δ) (M) × Jd−j (δ) (M)
(1 − t ) 2 2
(8.3)
t 2(2j (δ)−d+g−1)  d−j (δ) 
− Pt S M × Jj (δ) (M)
1−t 2
 Morse theory and stable pairs 175

(II+ ) These strata are indexed by integers j = j (δ) = deg # = deg L1 such
that d − N + 1 ≤ j ≤ N − 1, and δ = j − d + τ . The contribution is

t 2j (δ)  
II+
j (t) = Pt S j (δ) M × Jd−j (δ) (M) (8.4)
1−t 2

(II− ) These strata are indexed by integers j = d − j (δ) such that 0 ≤ j ≤

d − N, where δ = j (δ) − τ = d − j − τ , and the contribution is

t 2(2j (δ)−d+g−1)  d−j (δ) 

II−
j (t) = Pt S M × Jj (δ) (M) (8.5)
1 − t2
Then we have

Theorem 8.4.1 For τ ∈ (max{d/2, N − 1}, N),

Pt (Mτ,d ) = PtG (Bss

τ
)
N−1 ∞
j
= Pt (BG) − Id/2
a (t) − Ija (t) − Ib (t)
j =!d/2+1" j =N
d−N N−1
− II−
j (t) − II+
j (t)
j =0 j =d−N+1

Proof By Theorem 8.3.11 we have

PtG (Bss
τ
) = Pt (BG) − PtG (Xδ , Xδ1 )
δ∈τ,d \{0}

If δ ∈ +
τ,d ∩ (τ − d/2, τ ], then by the Morse–Bott lemma (8.26) and (8.29),

t 2σ (δ) G
PtG (Xδ , Xδ1 ) = P (ηδ )
1 − t2 t

where σ (δ) is given in Corollary 8.3.14. If δ ∈ +

τ,d ∩ (τ − d/2, τ ] then by
Section 8.3.6,

PtG (Xδ , Xδ1 ) = PtG (Xδ , Xδ ) − PtG (Xδ1 , Xδ )

The first term on the right hand side is given by (8.35). For the second term,
we have
6
∗  HG∗ (νI,δ
 
, νI,δ ) δ ∈ +
τ,d ∩ [2τ − d, τ ]
HG (Xδ1 , Xδ ) =
HG (ωδ , νI,δ ) δ ∈ +
∗ 
τ,d ∩ (τ − d/2, 2τ − d)
176 R. A. Wentworth and G. Wilkin

and the latter cohomology groups have been computed in (8.20) and (8.25).
This completes the computation. 

When the parameter τ is non-generic (i.e. τ = N for some integer N ∈

[d/2, d]) then the same analysis as above applies, however now there are
split solutions to the vortex equations. These correspond to one of the critical
sets of type II, where E = L1 ⊕ L2 with φ ∈ H 0 (L1 ) \ {0}, and deg L2 = τ .
Therefore, the only difference the generic and non-generic case is that we do
not count any contribution from the critical set of type II− with j = d − N .
Therefore the Poincaré polynomial is

Theorem 8.4.2 For τ = N ,

N−1 ∞
PtG (Bss
j
N
) = Pt (BG) − Id/2
a (t) − Ija (t) − Ib (t)
j =!d/2+1" j =N
d−N−1 N−1
− II−
j (t) − II+
j (t)
j =0 j =d−N+1

Finally, using Theorem 8.4.1, we can give a computation of the remaining

term which is as yet undetermined in low degree.

Lemma 8.4.3 For all d ≥ 2,

!d/2"
1 t 2j − t 2(d+g−1−2j )
a (t) =
Id/2 PtG (Ass ) − Pt (S j M × Jd−j (M))
1−t 2
j =0
1 − t 2

⎧
⎨0 if d odd
− t 2g−2
⎩ Pt (S d/2 M × Jd/2 (M)) if d even
(1 − t 2 )

Remark 8.4.4 It can be verified directly that for d > 4g − 4, the expression
above agrees with (8.1). See the argument of Zagier in [20, pp. 336–7].

Proof of Lemma 8.4.3 Take the special case N = d. Then Mτ,d is a projective
bundle over Jd (M), and so

1 − t 2(d+g−1)
Pt (Mτ,d ) = Pt (Jd (M))
1 − t2
 Morse theory and stable pairs 177

On the other hand, from Theorem 3.1 we have

d−1 ∞ d−1
Ib (t) − II− II+
j
Pt (Mτ,d ) = Pt (BG) − Id/2
a (t) − Ija (t) − 0 (t) − j (t)
j =!d/2+1" j =d j =1

Now notice that the term II− d

0 (t) is cancelled by the second term in Ib . We
j j
combine the remaining terms in the sum of Ib with the sum of Ia . We
have
∞
t 2(2j −d+g−1)
Pt (Mτ,d ) = Pt (BG) − Id/2
a (t) − Pt (Jj (M) × Jd−j (M))
j =!d/2+1"
(1 − t 2 )2

d−1
t 2j
+ P (S j M × Jd−j (M))
2) t
j =!d/2+1"
(1 − t

d−1
t 2(2j −d+g−1)
+ Pt (S d−j M × Jj (M))
j =!d/2+1"
(1 − t 2)

d−1
t 2j
− Pt (S j M × Jd−j (M))
j =1
(1 − t 2 )

∞
t 2(2j −d+g−1)
= Pt (BG) − Id/2
a (t) − Pt (Jj (M) × Jd−j (M))
j =!d/2+1"
(1 − t 2 )2

d−1
t 2(2j −d+g−1)
+ Pt (S d−j M × Jj (M))
j =!d/2+1"
(1 − t 2)

!d/2"
t 2j
− Pt (S j M × Jd−j (M))
j =1
(1 − t 2 )

Now make the substitution j → d − j in the second to the last sum, using
⎧
⎨d/2 − 1 = !d/2" − 1 if d even
d − !d/2 + 1" =
⎩d/2 − 1/2 = !d/2" if d odd

The result now follows from this, [1, Thm. 7.14], and the fact that Pt (Mτ,d ) is
equal to the j = 0 term in the sum. 
178 R. A. Wentworth and G. Wilkin

8.4.2 Comparison with the results of Thaddeus

In [20], Thaddeus computed the Poincaré polynomial of the moduli space
using different methods to those of this paper. The idea is to show that when
the parameter τ passes a critical value, then the moduli space Mτ,d undergoes
a birational transformation consisting of a blow-down along a submanifold and
a blow-up along a different submanifold (these transformations are known as
“flips”). By computing the change in Poincaré polynomial caused by the flips
as the parameter crosses the critical values, and also observing that the moduli
space is a projective space for one extreme value of τ , Thaddeus computed the
Poincaré polynomial of the moduli space for any value of the parameter. In this
section we recover this result from Theorem 8.4.1. In the Morse theory picture
we see that the critical point structure changes: As τ increases past a critical
value then a new critical set appears, and the index may change at existing
critical points.
Theorem 8.4.5 Let N ∈ Z, d/2 < N ≤ d − 1. Then for τ∈
(max(d/2, N − 1), N),
t 4N−2d+2g−2 − t 2d−2N  d−N 
Pt (Mτ +1,d ) − Pt (Mτ,d ) = Pt S M × JN (M)
1 − t2
(8.6)
As a consequence, the Poincaré polynomial of the moduli space has the
form
(1 + t)2g t 2d+2g−2−4N t 2N+2 (1 + xt)2g
Pt (Mτ,d ) = Coeffx N −
1 − t2 xt − 1
4 x − t2 (1 − x)(1 − xt 2 )
(8.7)
Remark 8.4.6 Let M0τ,d denote the moduli space where the bundle has fixed
determinant (see [20]). The analysis in this paper applies in this case as well.
In particular, one obtains
t 4N−2d+2g−2 − t 2d−2N  d−N 
Pt (M0τ +1,d ) − Pt (M0τ,d ) = Pt S M (8.8)
1 − t2
This exactly corresponds to Thaddeus’ results for Pt (PWj+ ) − Pt (PWj− ) [20,
p. 336], where j = d − N.

Proof of Theorem 8.4.5 By Theorem 8.4.1,

− + +
Pt (Mτ +1,d ) − Pt (Mτ,d ) = −IN
a (t) + Ib (t) + IId−N (t) − IId−N (t) − IIN (t)
N

(8.9)
 Morse theory and stable pairs 179

Substituting in the results of (8.2), (8.3), (8.5), and (8.4) gives

t 2(2N−d+g−1)
Pt (Mτ +1,d ) − Pt (Mτ,d ) = − Pt (JN (M) × Jd−N (M))
(1 − t 2 )2
t 2N  
+ Pt S N M × Jd−N (M)
1−t 2

t 2(2N−d+g−1)  d−N 
+ Pt S M × JN (M)
1 − t2
t 2(2N−d+g−1)
+ Pt (JN (M) × Jd−N (M))
(1 − t 2 )2
t 2(2N−d+g−1)  d−N 
− Pt S M × JN (M)
1−t 2

t 2(d−2(d−N)+g−1)  d−N 
+ Pt S M × JN (M)
1−t 2

t 2(d−N)  d−N 
− Pt S M × JN (M)
1−t 2

t 2N  
− Pt S N M × Jd−N (M)
1 − t2
1   
= Pt S d−N M ×JN (M) t 4N−2d+2g−2 −t 2d−2N
1 − t2
as required. Using the results of [16] on the cohomology of the symmetric
product, and the fact that Pt (JN (M)) = (1 + t)2g , we see that the same method
as for the proof of [20, (4.1)] gives equation (8.7). 

Remark 8.4.7 For τ as above, Theorem 8.4.2 shows that the difference
t 4N−2d+2g−2  d−N 
PtG (Bss
N
) − Pt (Mτ,d ) = II−
d−N (t) = Pt S M × JN (M)
1 − t2
comes from only one critical set; the type II critical set corresponding to a
solution of the vortex equations when τ = N . The rest of the terms in (8.9),
corresponding to the difference
+ +
Pt (Mτ +1,d ) − PtG (Bss
N
) = −IN
a (t) + Ib (t) − IId−N (t) − IIN (t)
N

t 2N  
=− Pt S N M × Jd−N (M)
1−t 2

come from a number of changes that occur in the structure of the critical
sets as τ increases past N : the term −II+
d−N (t) corresponds to the type II
180 R. A. Wentworth and G. Wilkin

critical point that no longer is a solution to the vortex equations, the term
−II+ +
N (t) corresponds to the new critical point of type II that appears, and the
term −Ia (t) + Ib (t) corresponds to the critical point that changes type from Ib
N N

to Ia .
Therefore we see that the changes in the critical set structure as τ crosses
the critical value N are localized to two regions of B. The first corresponds to
interchanging critical sets of type II− and type II+ . This is the phenomenon
illustrated in Figure 1. The second corresponds to critical sets of type Ia and
II+ that merge to form a single component of type Ib . The terms from the first
change exactly correspond to those in (8.6), i.e.

t 4N−2d+2g−2 − t 2d−2N  d−N 

II− +
d−N (t) − IId−N (t) = Pt S M × JN (M)
1−t 2

= Pt (Mτ +1,d ) − Pt (Mτ,d )

b (t) − Ia (t) −
and the terms from the second change cancel each other, i.e. IN N

II+
N (t) = 0.

References
[1] M. F. Atiyah and R. Bott. The Yang-Mills equations over Riemann surfaces. Philos.
Trans. Roy. Soc. London Ser. A, 308 (1983), 523–615.
[2] R. Bott. Nondegenerate critical manifolds. Ann. of Math., 60 (1954), no. 2, 248–
261.
[3] S. Bradlow. Special metrics and stability for holomorphic bundles with global
sections. J. Differential Geom. 33 (1991), no. 1, 169–213.
[4] S. Bradlow and G.D. Daskalopoulos. Moduli of stable pairs for holomorphic
bundles over Riemann surfaces. Internat. J. Math. 2 (1991), no. 5, 477–513.
[5] S. Bradlow, G.D. Daskalopoulos, and R.A. Wentworth. Birational equivalences of
vortex moduli. Topology 35 (1996), no. 3, 731–748.
[6] G.D. Daskalopoulos. The topology of the space of stable bundles on a compact
Riemann surface. J. Differential Geom. 36 (1992), no. 3, 699–746.
[7] G.D. Daskalopoulos and R.A. Wentworth. Convergence properties of the Yang-
Mills flow on Kähler surfaces. J. Reine Angew. Math. 575 (2004), 69–99.
[8] G.D. Daskalopoulos, R.A. Wentworth, J. Weitsman, and G. Wilkin. Morse theory
and hyperkähler Kirwan surjectivity for Higgs bundles. J. Differential Geom. 87
(2011), no. 1, 81–116.
[9] S. K. Donaldson. Anti self-dual Yang-Mills connections over complex algebraic
surfaces and stable vector bundles. Proc. London Math. Soc. (3) 50(1) (1985),
1–26.
[10] M.-C. Hong. Heat flow for the Yang-Mills-Higgs field and the Hermitian Yang-
Mills-Higgs metric. Ann. Global Anal. Geom. 20(1) (2001), 23–46.
 Morse theory and stable pairs 181

[11] M.-C. Hong and G. Tian. Asymptotical behaviour of the Yang-Mills flow and
singular Yang-Mills connections. Math. Ann. 330(3) (2004), 441–472.
[12] A. King and P. Newstead. Moduli of Brill-Noether pairs on algebraic curves.
Internat. J. Math. 6 (1995), no. 5, 733–748.
[13] F. C. Kirwan. “Cohomology of quotients in symplectic and algebraic geometry”,
volume 31 of Mathematical Notes. Princeton University Press, Princeton, NJ,
1984.
[14] S. Kobayashi. “Differential geometry of complex vector bundles”, volume 15 of
Publications of the Mathematical Society of Japan. Princeton University Press,
Princeton, NJ, 1987. , Kano Memorial Lectures, 5.
[15] J. Le Potier. Systèmes cohérents et structures de niveau. Astérisque No. 214 (1993),
143 pp.
[16] I. G. Macdonald. Symmetric products of an algebraic curve. Topology 1 (1962),
319–343.
[17] J. Råde. On the Yang-Mills heat equation in two and three dimensions. J. Reine
Angew. Math. 431 (1992), 123–163.
[18] N. Raghavendra and P. Vishwanath. Moduli of pairs and generalized theta divisors.
Tohoku Math. J. (2) 46 (1994), no. 3, 321–340.
[19] L. Simon. Asymptotics for a class of nonlinear evolution equations, with applica-
tions to geometric problems. Ann. of Math. (2) 118(3) (1983), 525–571.
[20] M. Thaddeus. Stable pairs, linear systems and the Verlinde formula. Invent. Math.
117, no. 2 (1994), 317–353.
[21] G. Wilkin. Morse theory for the space of Higgs bundles. Comm. Anal.Geom. 16,
no. 2 (2008), 283–332.

Authors’ addresses:

Richard A. Wentworth
Department of Mathematics,
University of Maryland,
College Park, MD 20742, USA
[email protected]

Graeme Wilkin
Department of Mathematics
University of Colorado
Boulder, CO 80309
USA
[email protected]
 9
Manifolds with k-positive Ricci curvature
jon wolfson

9.1 Introduction
Let (M, g) be an n-dimensional Riemannian manifold. We say M has k-positive
Ricci curvature if at each point p ∈ M the sum of the k smallest eigenvalues of
the Ricci curvature at p is positive. We say that the k-positive Ricci curvature
is bounded below by α if the sum of the k smallest eigenvalues is greater than
α. Note that n-positive Ricci curvature is equivalent to positive scalar curvature
and one-positive Ricci curvature is equivalent to positive Ricci curvature. We
first describe some basic connect sum and surgery results for k-positive Ricci
curvature that are direct generalizations of the well known results for positive
scalar curvature (n-positive Ricci curvature). Using these results we construct
examples that motivate questions and conjectures in the cases of 2-positive and
(n − 1)-positive Ricci curvature. In particular:

Conjecture 1 If M is a closed n-manifold that admits a metric with 2-positive

Ricci curvature then the fundamental group, π1 (M), is virtually free.

We formulate an approach to solving this conjecture based, at least philo-

sophically, on the method used in the proof of the Bonnet–Myers theorem:
A closed n-manifold that admits a metric with positive Ricci curvature (1-
positive Ricci curvature) has finite fundamental group. The Bonnet-Myers
theorem proves a diameter bound for a manifold with positive Ricci curvature
bounded below and then uses covering spaces to conclude the result on the
fundamental group. In our approach to Conjecture 1 we describe a suitable
notion of “two-dimensional diameter bound”, namely, the notion of fill radius.
This idea was introduced in [G], [G-L2], [S-Y2]. We show that if a closed
manifold satisfies a curvature condition that implies a fill radius bound then the
fundamental group is virtually free. The proof of this statement uses covering

182
 Manifolds with k-positive Ricci curvature 183

space theory, like in the Bonnet–Myers argument, as well as some notions from
geometric group theory. See [R-W] for full details and Section 4 of this chap-
ter for an overview. However, the question: “Does a manifold with 2-positive
Ricci curvature bounded below by α satisfy an upper bound on its fill radius?”
remains open.
It is clear that k-positive Ricci curvature implies (k + 1)-positive Ricci
curvature, so results on n-manifolds with n-positive Ricci curvature hold for n-
manifolds with k-positive Ricci curvature, any k, 0 < k < n. As a result of the
surgery theorem we pose some interesting questions on the difference between
n-positive Ricci curvature and (n − 1)-positive Ricci curvature.
The author was partially supported by NSF grant DMS-0604759.

9.2 Manifolds with k-positive Ricci curvature

In this section we state some structure theorems for manifolds with k-positive
Ricci curvature and provide some examples. Throughout we assume that (M, g)
is an n-dimensional Riemannian manifold with n ≥ 3. Recall that n-positive
Ricci curvature is positive scalar curvature and one-positive Ricci curvature
is positive Ricci curvature. The following is a direct generalization of a well-
known result of Gromov–Lawson [G-L1] and Schoen–Yau [S-Y1] on con-
nect sum and surgeries of manifolds with positive scalar curvature (also see
[R-S]).
Theorem 9.2.1 Let M be a compact n-manifold with a metric of k-positive
Ricci curvature, 2 ≤ k ≤ n. Then any manifold obtained from M by performing
surgeries in codimension q with q ≥ max{n + 2 − k, 3} also has a metric of
k-positive Ricci curvature. If M1 and M2 are compact n-manifolds with metrics
of k-positive Ricci curvature, 2 ≤ k ≤ n, then their connect sum M1 #M2 also
carries a metric with k-positive Ricci curvature.
Proof The proof follows easily from the procedure provided in the proof
in [G-L1] of the similar statement for positive scalar curvature (the case of
k = n). We note that this procedure fails when k = 1 (the case of positive Ricci
curvature). For the sake of completeness we will give the proof here of the
connect sum result following, sometimes verbatim, the method of Gromov–
Lawson. Suppose that M is an n-manifold, n ≥ 3, and that M has a metric with
Ricci curvature that is k-positive, for 2 ≤ k ≤ n (i.e., at each point the sum of
any k eigenvalues of the Ricci curvature is positive). Fix a point p ∈ M and
consider a normal coordinate ball D centered at p. Following [G-L1] we will
184 J. Wolfson

change the metric in D preserving that the Ricci curvature is k-positive such
that the metric agrees with the original metric near ∂D and such that near p
the metric is the standard metric on S (n−1) (ε) × R, for any ε sufficiently small.
It follows from this that 1-handles can be added and connected sums taken
preserving that the Ricci curvature is k-positive, for any 2 ≤ k ≤ n.
The method of Gromov–Lawson proceeds as follows: We consider the Rie-
mannian product D × R with coordinates (x, t), where x are normal coordinates
on D. We define a hypersurface N ⊂ D × R by the relation

N = {(x, t) : (x, t) ∈ γ }

where γ is a smooth curve in the (r, t)-plane that is monotonically decreasing,

begins along the positive r-axis and ends as a straight line parallel to the t-
axis. The metric induced on N from D × R extends the metric on D near its
boundary and ends with a metric on S n−1 (ε) × R. If ε is sufficiently small then
by Lemma 1 of [G-L1] we can change the metric in this tubular piece to the
product metric on the product of the standard ε-sphere with R.
The key problem is to choose the curve γ so that the metric induced on N
has strictly k-positive Ricci curvature at all points. To do this we begin as in
[G-L1] by letting  be a geodesic ray in D beginning at the origin. Then the
surface  × R is totally geodesic in D × R and the normal to N along points
of N ∩ ( × R) lies in  × R. It follows that γ = N ∩ ( × R) is a principal
curve on N and that the associated principal curvature at a point (r, t) ∈ γ is
the curvature κ of γ at that point. The remaining principal curvatures at such a
point are of the form (−1/r + O(r)) × sin θ where θ is the angle between the
normal to the hypersurface and the t-axis.
Fix a point q ∈ γ ⊂ N corresponding to a point (r, t) ∈ γ . Let {e1 , . . . , en }
be an orthonormal basis of Tq (N) such that e1 is the tangent vector to γ
and {e2 , . . . , en } (which are tangent vectors to D) are principal vectors for
the second fundamental form of N . The Gauss curvature equation relates the
curvature tensor Rij lm of N with the curvature tensor R ij lm of D × R. In
particular, with respect to this basis, at q:

R ij ij = Rij ij − λi λj for i = j
R ij lj = Rij lj for i = l.

where λ1 , . . . , λn are the principal curvatures corresponding to the directions

e1 , . . . , en respectively. As above, λ1 = κ, the curvature of γ in  × R and
λj = (−1/r + O(r)) sin θ for j = 2, . . . n. Since D × R has the product metric
 Manifolds with k-positive Ricci curvature 185

we have:

R ij lj = RijDlj , for i, j, l = 2, . . . , n,
R 1j ij = R D∂ j ij cos θ, for i, j = 2, . . . , n,
∂r
R i1j 1 = RiD∂ j ∂ cos2 θ, for i, j = 2, . . . , n,
∂r ∂r

where R D is the curvature tensor of the metric on D. It follows that the Ricci
curvature of N at (x, t) with respect to the frame {e1 , . . . , en } is given by:
n
Ric11 = RicD ∂
, ∂
∂r ∂r
cos2 θ + κ λj (9.2.1)
j =2

Ric1j = RicD ∂
∂r
, ej cos θ, for j = 2, . . . , n,
Ricij = RicD
ij −RiD∂ j ∂ sin θ, 2
for i, j = 2, . . . , n, i = j,
∂r ∂r
⎛ ⎞
n
Ricii = RicD
ii −Ri ∂ i ∂ sin θ + λi
D 2 ⎝ λj + κ ⎠ , for i = 2, . . . , n,
∂r ∂r
j =2,j =i

where RicD denotes the Ricci curvature of D.

The eigenvalues of a matrix depend continuously on the entries of the matrix.
Therefore for θ sufficiently small, using that λi = (−1/r + O(r)) sin θ , for
i ≥ 2, the Ricci curvature at (x, t) in formula (9.2.1) is k-positive. In particular,
there is an θ0 > 0 such that for 0 ≤ θ ≤ θ0 the Ricci curvature at (x, t) is k-
positive. As in [G-L1] we bend γ to the angle θ0 and continue γ as a straight
line segment. Denote the straight line segment of γ by γ0 . Along this curve N
has Ricci curvature that is k-positive. Since κ ≡ 0 along γ0 , we see that as r
becomes small the Ricci curvature of N is of the form:

Ricij = O(1), for i = j, (9.2.2)

(n − 2) sin θ0 2
Ricii = + O(1), for i = 2, . . . , n,
r2
Ric11 = O(1).

Choose r0 > 0 small. From (9.2.2) it follows that the the Ricci curvature of the
hypersurface N has (n − 1) positive eigenvalues and that each of these eigenval-
ues is larger in absolute value than the one remaining eigenvalue (corresponding
to the direction of e1 ). Therefore the Ricci curvature remains k-positive and, in
fact, becomes 2-positive. Consider the point (r0 , t0 ) ∈ γ0 . Bend γ0 , beginning
at this point, with the curvature function κ(s) similar to the one used in [G-L1]:
186 J. Wolfson

κ
sin θ0
2r0

s
1
2 r0

Here s is the arclength parameter. Since max κ = sin θ0

2r0
we see that:
⎛ ⎞  
 n 
n
 
λi ⎝ λj + κ ⎠ > κ λj  > 0, for i = 2, . . . , n.
j =2,j =i  j =2 

Using (9.2.1) (and r0 sufficiently small) this implies that along the bend the
Ricci curvature of N remains 2-positive. During this bending process the curve
will not cross the line r = r0 /2 since the length of the bend is r0 /2 and it
begins at height r0 . The total amount of bending is:

sin θ0
θ = κds  ,
4
independent of r0 .
Continue the curve with a new straight line segment γ1 at an angle θ1 =
θ0 + θ . Repeat the above procedure now using θ1 where previously we used
θ0 . The total bending of this procedure will then be sin4θ1 > sin4θ0 . By repeating
this procedure a finite number of times (depending on sin θ0 ) we can achieve a
total bend of π/2. This completes the proof of the connect sum result.
For the general case of surgeries we again explain the modification of the
Gromov-Lawson argument. Let S p ⊂ M be an embedded sphere with trivial
normal bundle B of dimension q ≥ 3. Identify B with S p × Rq . Define r :
S p × Rq → R+ by r(y, x) = ||x||, and set S p × D q (ρ) = {(y, x) : r(y, x) ≤
ρ}. Choose r̄ > 0 such that the normal exponential map exp : B → M is an
embedding on S p × D q (r̄) ⊂ B. Lift the metric on M to S p × D q (r̄) by the
exponential map. Then r is the distance function to S p × {0} in S p × D q (r̄),
and curves of the form {y} × , where  is a ray in D q (r̄) emanating from the
origin, are geodesics in S p × D q (r̄).
We now consider hypersurfaces in the Riemannian product (S p × D q (r̄)) ×
R of the form:

N = {(y, x, t) : (r(y, x), t) ∈ γ }

 Manifolds with k-positive Ricci curvature 187

where γ is, as above, a curve in the (r, t)-plane. As in the connect sum case
we must choose γ so that the metric on N has k-positive Ricci curvature at all
points. We first remark that γ = N ∩ ({y} ×  × R) is a principal curve on N
and the associated principal curvature at a point corresponding to (r, t) ∈ γ is
exactly the curvature κ of γ at that point.
Now fix a point q ∈ γ ⊂ N corresponding to a point (r, t) ∈ γ . Let
{e1 , . . . , en } be an orthonormal basis of Tq (N ) such that e1 is the tangent
vector to γ , and e2 , . . . , en are principal vectors for the second fundamen-
tal form of N . If the metric on S p × D q (r̄) at q is the product metric
then the vectors e2 , . . . , eq can be chosen to be tangent to {y} × D q (r̄) and
the vectors eq+1 , . . . , en to be tangent to S p × {x}. The principal curvatures
λ2 , . . . , λq are then of the form (− 1r + O(r)) sin θ , where θ is the angle
between the normal to the hypersurface and the t-axis and the principal cur-
vatures λq+1 , . . . , λn are of the form O(1) sin θ (i.e., are independent of r).
In the general case such a simple description is not possible. However if r is
small the q − 1 largest principal curvatures are of the form (− rc + O(1)) sin θ ,
where c is a positive constant that can be bounded away from zero and the
remaining principal curvatures are of this form O(1) sin θ . (Since the prod-
uct of the principal curvatures grows like r q−1 1
.) We will denote the q − 1
largest principal curvatures by λ2 , . . . , λq corresponding to the directions
e2 , . . . , eq and the remainder by λq+1 , . . . , λn corresponding to the directions
eq+1 , . . . , en .
The Gauss curvature equation relates the curvature tensor Rij lm of N with
the curvature tensor R ij lm of (S p × D q (r̄)) × R. In particular, with respect to
this basis, at q:

R ij ij = Rij ij − λi λj , for i = j,
R ij lj = Rij lj , for i = l,

where λ1 , . . . , λn are the principal curvatures corresponding to the directions

e1 , . . . , en respectively. As above, λ1 = κ, the curvature of γ in  × R and
λj = (− rc + O(1)) sin θ for j = 2, . . . , q. The remaining principal curvatures
λq+1 , . . . , λn are of the form O(1) sin θ . For this reason, unlike in the connect
sum case, they play no useful role in the following computation. Since (S p ×
D q (r̄)) × R has the product metric in the second factor we have:

R ij lj = RijS lj×D ,
p q
for i, j, l = 2, . . . , n,
R 1j ij = R S∂ j×D
p q

ij
cos θ, for i, j = 2, . . . , n,
∂rp
R i1j 1 = RiS ∂ ×D
q

j ∂
cos2 θ, for i, j = 2, . . . , n,
∂r ∂r
188 J. Wolfson

where R S ×D is the curvature tensor of the metric on S p × D q . It follows that

p q

the Ricci curvature of N at (y, x, t) with respect to the frame {e1 , . . . , en } is

given by:
n
p
×D q
Ric11 = RicS ∂
, ∂
∂r ∂r
cos2 θ + κ λj (9.2.3)
j =2
p
×D q
Ric1j = RicS ∂
∂r
, ej cos θ, for j = 2, . . . , n,
×D
−RiS ∂ ×D
p q p q
Ricij = RicSij j ∂
sin2 θ, for i, j = 2, . . . , n, i = j,
∂r ∂r
⎛ ⎞
n
×D q
−RiS ∂ ×D sin2 θ + λi ⎝ λj + κ ⎠ , for i = 2, . . . , n,
p p q
Ricii = RicSii i∂
∂r ∂r
j =2,j =i

where RicS ×D denotes the Ricci curvature of S p × D q .

p q

Since λi = (− rc + O(1)) sin θ , for i = 2, . . . , q and λi = O(1) sin θ , for

i = q + 1, . . . , n, if θ is sufficiently small, the Ricci curvature at (y, x, t) in
formula (9.2.3) is k-positive. In particular, there is an θ0 > 0 such that for
0 ≤ θ ≤ θ0 the Ricci curvature at (y, x, t) is k-positive. As above we bend γ to
the angle θ0 and continue γ as a straight line segment. Denote the straight line
segment of γ by γ0 . Along this curve N has Ricci curvature that is k-positive.
Since κ ≡ 0 along γ0 , we see that as r becomes small the Ricci curvature of N
is of the form:

Ricij = O(1), for i = j, (9.2.4)

(q − 2)c2 sin2 θ0
Ricii = +O 1
r
, for i = 2, . . . , q,
r2
Ricii = O 1
r
, for i = q + 1, . . . , n,
Ric11 = O(1).

(Note that since λq+1 , . . . , λn are bounded for r small, we cannot conclude any
more than Ricii = O( 1r ) for i = q + 1, . . . , n.) Therefore, provided q ≥ 3, for
r sufficiently small the Ricci curvature has q − 1 positive eigenvalues that
strongly dominate all other eigenvalues. In particular, provided k > n − q + 1
the Ricci curvature remains k-positive. We can then bend γ to a line parallel to
the t-axis and preserve k-positive Ricci curvature as above.
This construction of N determines a “tube” of k-positive Ricci curvature
with two boundary components. The initial boundary component has a collar
neighborhood isometric to a tubular neighborhood of S p in M. The final bound-
ary component has a collar neighborhood isometric to ∂(S p × D q (ε)) × R =
S p × S q−1 (ε) × R for the product metric with the R factor. This allows us to
 Manifolds with k-positive Ricci curvature 189

glue the initial end of N to M \ (S p × D q (r)) to construct a manifold M  with

k-positive Ricci curvature and with one boundary component S p × S q−1 (ε).
As in [G-L1] we next observe that the metric on ∂(S p × D q (ε)) = S p ×
S q−1 (ε) can be homotoped through metrics with k-positive Ricci curvature
to the standard product metric of euclidean spheres on S p × S q−1 (ε). The
argument that accomplishes this is the same as that in Gromov–Lawson: For
ε sufficiently small the metric on ∂(S p × D q (ε)) can be homotoped through
metrics with k-positive Ricci curvature to a metric on S p × S q−1 (ε) that is
a Riemannian submersion with totally geodesic fibers that have the euclidean
metric of curvature ε12 . When ε sufficiently small the terms Ric(v, v) for vertical
vectors v strongly dominate all other terms of the Ricci curvature and therefore
we can deform this metric, preserving k-positive Ricci curvature (for k >
n − q + 1), through Riemannian submersions to one with the standard metric
on S p .
Denote the induced metric on S p × S q−1 (ε) by g0 and the product of the
standard metrics on S p × S q−1 (ε) by g1 . The final step of the Gromov-Lawson
argument shows that the homotopy constructed in the previous paragraph can be
used to find a metric of k-positive Ricci curvature on S p × S q−1 (ε) × [0, a], for
some a > 0, whose restriction to a collar neighborhood of S p × S q−1 (ε) × {0}
is g0 + dt 2 and whose restriction to a collar neighborhood of S p × S q−1 (ε) ×
{a} is g1 + dt 2 . We will give a proof below in Proposition 9.2.2. Using this result
we can assume that the end of the manifold M  is isometric to S p × S q−1 (ε) × R
equipped with the product of the standard metrics on the spheres. From this the
surgery result follows immediately. 

To complete the proof of Theorem 9.2.1 we prove the following proposition

motivated by Proposition 3.3 of [R-S], Lemma 3 of [G-L] and [Ga].

Proposition 9.2.2 Let X be a compact n-manifold. Suppose that there is

a smooth family of Riemannian metrics on X, {gt }, 0 ≤ t ≤ 1, each with k-
positive Ricci curvature for some k, 2 ≤ k ≤ n. Then there is a Riemannian
metric g on X × [0, a], for some a > 0, with k-positive Ricci curvature such
that the restriction of g on a collar neighborhood of X × {0} is g0 + dt 2 and
the restriction of g on a collar neighborhood of X × {a} is g1 + dt 2 .

Proof Consider the metric gf (t) + dt 2 on X × [0, a], where f (t) is a C 2 ([0, a])
function that is monotonically increasing from 0 to 1. We will explic-
itly determine f (t) below. We compute the curvature of gf (t) + dt 2 at the
point (x0 , t0 ) ∈ X × [0, a]. Let e0 be the unit normal along the hypersur-
face X × {t0 } pointing in the direction ∂t∂ . Let {e1 , . . . , en } be an orthonor-
mal frame along X × {t0 } near (x0 , t0 ). Then {e0 , . . . , en } is an orthonormal
190 J. Wolfson

frame along X × [0, 1] near (x0 , t0 ). Let {ω0 , . . . , ωn } be the dual coframe.
For notational convenience we use the index ranges α, β, γ = 0, 1, . . . , n and
i, j, k = 1, 2, . . . , n. The connection one-form for the coframe {ω0 , . . . , ωn }
is {ωαβ }. The one-forms ωij depend only on the metric gf (t) . However the
one-forms ω0i depend on f  (t) so at (x0 , t0 ) we have:
ω0i = O(|f  |), (9.2.5)
   2
dω0i = O(|f |) + O(|f |) + O(|f | ).
The curvature two-form αβ on X × [0, a] is determined by the structure
equation:
αβ = dωαβ − ωαγ ∧ ωγβ . (9.2.6)
γ

Denote the curvature tensor of gf (t) + dt 2 on X × [0, a] by R̄αβγ δ and the

curvature tensor of gf (t0 ) on X × {t0 } by Rij kl . Then by the structure equation
(9.2.6) (or by the Gauss equation) at (x0 , t0 ):
R̄ij kl = Rij kl + O(|f  |2 ). (9.2.7)
By the structure equation and (9.2.5)
R̄0j kl = O(|f  |) + O(|f  |) + O(|f  |2 ),
R̄0j 0l = O(|f  |) + O(|f  |) + O(|f  |2 ).
Denote the Ricci curvature of gf (t) + dt 2 on X × [0, a] by Ric and the Ricci
curvature of gf (t0 ) on X × {t0 } by Ric. Then
Ricij = Ricij + O(|f  |) + O(|f  |) + O(|f  |2 ),
Rici0 = O(|f  |) + O(|f  |) + O(|f  |2 ), (9.2.8)
   2
Ric00 = O(|f |) + O(|f |) + O(|f | ).
Given constants, 0 < ε, λ ( 1, choose a > 0 sufficiently large so that there
is a C 2 function f : [0, a] → [0, 1] satisfying:
(i) f is monotonically increasing,
(ii) f (t) = 0, for 0 ≤ t ≤ ε,
(iii) f (t) = 1, for a − ε ≤ t ≤ a,
(iv) |f  (t)|, |f  (t)| < λ, for all t.
Using (9.2.8) since (X, gt ) has k-positive Ricci curvature for all t ∈ [0, 1], the
compactness of X and [0, 1] allow that λ can be chosen sufficiently small to
ensure that the metric gf (t) + dt 2 (for f as given above) has k-positive Ricci
curvature everywhere on X × [0, a]. The result follows. 
 Manifolds with k-positive Ricci curvature 191

Let X be a compact Riemannian manifold with positive Ricci curvature.

Then the manifold X × S 1 has a metric of 2-positive Ricci curvature (and of
non-negative Ricci curvature). Therefore, by Theorem 9.2.1, if Xi , i = 1, . . . , l,
are compact Riemannian (n − 1)-manifolds with positive Ricci curvature, the
manifolds:

#li (Xi × S 1 ), l ≥ 2, (9.2.9)

admit metrics with 2-positive Ricci curvature. From this we see that compact
manifolds with 2-positive Ricci curvature can have large fundamental groups.
In fact the fundamental groups of the examples (9.2.9) are virtually free. This
implies that the manifolds (9.2.9), when l ≥ 2, do not admit metrics of non-
negative Ricci curvature since the universal covers of (9.2.9) have infinitely
many ends. In contrast the universal cover of a compact manifold with non-
negative Ricci curvature splits isometrically as a product N × Rp , where N is a
compact manifold, and therefore has one or two ends. Many other topologically
distinct examples of compact manifolds that admit metrics of 2-positive Ricci
curvature can be constructed by taking the connect sum of manifolds of positive
Ricci curvature with the manifolds of (9.2.9).
Consider the round metric on the sphere of radius r, Srn−2 , and the hyperbolic
metric on the Riemann surface g of genus g ≥ 2. When r is sufficiently small
the manifolds Srn−2 × g admit metrics of 3-positive Ricci curvature but not
of 2-positive Ricci curvature. Examples of this type indicate that 3-positive
Ricci curvature when n ≥ 4 imposes much weaker restrictions on π1 (M) than
2-positive Ricci curvature.
The surgery results of Theorem 9.2.1 do not allow any surgeries preserving
2-positive Ricci curvature (except connect sum). When k > 2 and n > 3, q =
n − 1-surgeries preserve the curvature condition. This suggests a difference
between k = 2 and k > 2 in the rigidity of the fundamental group for manifolds
with k-positive Ricci curvature.

Conjecture 1 If M is a closed n-manifold that admits a metric with 2-positive

Ricci curvature then its fundamental group, π1 (M), is virtually free.

The first interesting case of this conjecture occurs when n = 4. When n = 3

the condition of positive scalar curvature is strictly weaker than 2-positive Ricci
curvature. In the case of positive scalar curvature the conjecture can be answered
in the affirmative. In fact, by the results of [S-Y1] and [G-L2] much more can
be said about the topology of 3-manifolds of positive scalar curvature. We
remark that the conjecture requires a positive curvature assumption. Requiring
only 2-non-negative Ricci curvature is not sufficient. For example that manifold
192 J. Wolfson

S 2 × T 2 , where S 2 is the round 2-sphere and T 2 is the flat 2-torus, has 2-non-
negative Ricci curvature and fundamental group isomorphic to Z ⊕ Z.
The surgery statement of Theorem 9.2.1 shows that if k = n − 1 then surgery
is possible provided q ≥ 3. This is the same condition as in the positive scalar
curvature case. The surgery result is used by Gromov-Lawson [G-L1] to prove
that if n ≥ 5 then every compact simply-connected non-spin n-manifold carries
a metric of positive scalar curvature and by Stolz [Sz] to prove that for n ≥ 5
every compact simply-connected spin n-manifold with vanishing α invariant
carries a metric of positive scalar curvature. Gromov-Lawson prove their result
for non-spin manifolds using oriented bordism. Stolz proves his result for spin
manifolds using spin bordism. Since n − 1 positive Ricci curvature implies
positive scalar curvature any necessary condition for positive scalar curvature
is a necessary condition for n − 1 positive Ricci curvature. In light of the
surgery result for n − 1 positive Ricci curvature exactly the same arguments
apply to prove:

Theorem 9.2.3 Let n ≥ 5. Every compact simply-connected non-spin n-

manifold carries a metric with n − 1 positive Ricci curvature. Every compact
simply-connected spin n-manifold with vanishing α invariant carries a metric
with n − 1 positive Ricci curvature.

To prove the theorem, all that needs to be checked is that the generators of
oriented bordism described in [G-L1] and the HP2 bundles used in [Sz] admit
metrics with n − 1 positive Ricci curvature. In particular for compact simply-
connected n-manifold with n ≥ 5 there is no distinction between positive scalar
curvature and n − 1 positive Ricci curvature.
Question 1 Is there a compact n-manifold, n ≥ 5, that admits a metric of
positive scalar curvature but does not admit a metric with n − 1 positive Ricci
curvature?

9.3 Fill radius and an approach to Conjecture 1

In this section we will describe an approach to Conjecture 1. In the Introduction
we recalled that the Bonnet-Myers theorem on the fundamental group of a
closed manifold with positive Ricci curvature is proved by deriving a diameter
bound on such manifolds. The 2-positive Ricci curvature condition implies a
positive lower bound on the sum of any two eigenvalues of Ricci. Accordingly
we seek to derive a “two-dimensional diameter bound” for such manifolds. The
notion we use for “two-dimensional diameter bound” is that of fill radius [G],
[G-L2], [S-Y2].
 Manifolds with k-positive Ricci curvature 193

Let γ be a smooth simple closed curve in M which bounds a disk in M. Set

Nr (γ ) = {x ∈ M : d(x, γ ) ≤ r}. We define the fill radius of γ to be:
fillrad(γ ) = sup{r : dist(γ , ∂M) > r and γ does not bound a disc in Nr (γ )}
We say a Riemannian manifold (M, g) has its fill radius bounded by C if every
smooth simple closed curve γ which bounds a disk in M satisfies,
fillrad(γ ) ≤ C.
Clearly if the diameter of (M, g) is bounded so is its fill radius. In particular,
if for all p ∈ M each eigenvalue λi , i = 1, . . . , n of Ric(p) satisfies λi ≥ α,
where α is a positive constant, then there is a constant C = C(α) such that the
fill radius of M is bounded by C.
In [G-L2] and [S-Y2] versions of the following result on positive scalar
curvature and fill radius are proved.
Theorem 9.3.1 Let (M, g) be a complete Riemannian three dimensional man-
ifold with bounded geometry and with positive scalar curvature S that satisfies
S ≥ α > 0, for a constant α. Then if γ is a smooth simple closed curve in M
which bounds a disk in M:
%
8 π
fillrad(γ ) ≤ √ .
3 α
The statements of the results and the details of the proofs in [G-L2] and
[S-Y2] differ but the essential ideas of the proof are the same. Initially observe
that if γ is a simple closed curve in M that bounds a disc  (more generally,
 can be taken to be a Riemann surface with boundary) then:
fillrad(γ ) ≤ sup d(x, γ ),
x∈

where d(−, −) is the distance in M. Clearly,

sup d(x, γ ) ≤ sup d (x, γ ),
x∈ x∈

where d (−, −) denotes the distance on  in the induced metric. Thus the
fill radius of γ can be bounded above by an upper bound on the diameter
of  in the induced metric. For arbitrary  spanning γ such a bound is, of
course, impossible. However in [G-L2] and [S-Y2],  is taken to be an area
minimizer among discs spanning γ (i.e., a solution of the Plateau problem). In
particular,  is a stable, minimal immersion. After perturbing γ inward along
 the minimal surface is strictly stable for normal variations vanishing on the
boundary. In [S-Y1] the second variation of area for a minimal surface  in a
three manifold M is given. Let f ν be a compactly supported normal variation,
194 J. Wolfson

where f ∈ C0∞ () and ν is a unit normal. Then the second variation formula
can be written:

d 2 || 1
2
= |∇f |2 + f 2 K − S − ||A||2 da. (9.3.1)
dt |t=0  2

Here || = Area(), S is the scalar curvature of M, K is the Gauss curvature

of  and A is the second fundamental form. It follows then that if  is a strictly
stable minimal surface the second order linear elliptic operator:

L(f ) = f + f (K − S) (9.3.2)

is a positive operator on f ∈ C0∞ (). Denote the first eigenfunction of (9.3.2)

for the Dirichlet problem by k. Then k is positive on  and vanishes on the
boundary. It is the positivity of (9.3.2) that is used in both [G-L2] and [S-Y2]
to derive a diameter bound on . In [S-Y2] the argument proceeds as follows.
The argument in [G-L2] is somewhat different.
Let  be a strictly stable minimal surface with boundary ∂. Fix x ∈  and
consider the familyF of curves on  from x to ∂. For σ in F consider the
functional F (σ ) = σ k(σ (t))|σ  (t)|dt, where k is the eigenfunction of (9.3.2).
Minimize F over F. A smooth minima can be found by the direct method
(as with geodesics). Denote the minimizer by τ . Since F (σ ) is invariant under
reparameterization of the curve σ , we can suppose that τ is parameterized by
arclength. Set the length of τ to be . Since τ is a minimizer it is stable under
normal variations fixing the end points. Let μ be a unit normal vector field along
τ in . The normal vector field X = ψμ, where ψ is a function of the arclength
parameter s that vanishes at its endpoints, is an admissible variational vector
field. The second variation of the functional F at τ determines a quadratic form
I given by:

I (ψ, ψ) =
 
dk d 2k 2
− ψ  +k −1 ψ  +ψ(K +k −1 k +k −1 2 ) ψ + 2 (∇k · ν)2 ψ 2 kds
0 ds ds k
(9.3.3)

where  is the Laplace-Beltrami operator on  and K is the Gauss curvature

of . (This expression uses that ψ vanishes at its endpoints. In the general case
there is a boundary term.) Set
/ 2 0
 −1  dk −1 −1 d k
L0 (ψ) = − ψ + k ψ + ψ K + k k + k . (9.3.4)
ds ds 2
 Manifolds with k-positive Ricci curvature 195

Then,
 
2
I (ψ, ψ) = L0 (ψ)ψ − (∇k · ν)2 ψ 2 kds (9.3.5)
0 k2

On a minimizer I (ψ, ψ) ≥ 0 and hence, since the term 2

k2
(∇k · ν)2 ψ 2 ≥ 0,
 
L0 (ψ)ψkds ≥ 0 (9.3.6)
0

for all functions ψ on [0, ] vanishing at the endpoints.

Using that k is the first eigenfunction of L (9.3.4) becomes:
/ 0
dk d 2k
L0 (ψ) = − ψ  + k −1 ψ  + ψ k −1 L(k) + S + k −1 2 . (9.3.7)
ds ds
Choose a function g ∈ C ∞ ([0, ]) that is a first eigenfunction of L0 on [0, ]
for the Dirichlet problem. Then g > 0 on (0, ), g vanishes at the endpoints
and L0 (g) ≥ 0. Hence,

g  + k −1 g  k  + g(k −1 L(k) + S + k −1 k  ) ≤ 0.

Since k −1 L(k) > 0, this implies,

g −1 g  + g −1 k −1 g  k  + S + k −1 k  ≤ 0, (9.3.8)

on [0, ]. Let ϕ be any smooth function on [0, ] vanishing at the endpoints and
multiply (9.3.8) by ϕ 2 to give:
 
(g −1 g  ϕ 2 + g −1 k −1 g  k  ϕ 2 + k −1 k  ϕ 2 + Sϕ 2 )ds ≤ 0
0

Integrate by parts to give,

  2
d
1
2
[(g −1 g  )2 + (k −1 k  )2 ]ϕ 2 + 1
2
ln(gk) ϕ 2 + Sϕ 2 ds
0 dt
 
d
≤2 ϕϕ  ln(gk) ds (9.3.9)
0 dt
Note that,
 
  2
2ϕϕ  d ln(gk)  ≤ 4 (ϕ  )2 + 3 ϕ 2 d
ln(gk)
 dt  3 4
dt
2
d
≤ 43 (ϕ  )2 + 12 ϕ 2 ((g −1 g  )2 + (k −1 k  )2 ) + 12 ϕ 2 ln(gk) ,
dt
(9.3.10)
196 J. Wolfson

where the second inequality follows from,

2
d
1
4
ln(gk) ≤ 12 ((g −1 g  )2 + (k −1 k  )2 )
dt
Combining (9.3.9) and (9.3.10) we have:
   
1
2
α ϕ 2
ds ≤ 4
3
(ϕ  )2 ds, (9.3.11)
0 0

where 0 < α ≤ S. Thus,

 
− ϕ  − 38 αϕ ϕds ≥ 0,
0

and so the operator:

d2 3
− 2
− α
ds 8
has nonnegative first eigenvalue on [0, ]. Hence,
%
8 π
≤ √
3 α
This inequality holds for any x ∈ . Therefore,

sup dist (x, ∂) ≤ .

x∈

From this the theorem follows.

A fill radius bound has strong geometric implications. To illustrate this we
describe two results. The first, due to [S-Y1], [G-L2], concerns closed three
manifolds with positive scalar curvature. According to Milnor [Mi] any closed
three manifold M has a connect sum decomposition:

M = S1 # . . . #Sk #(S 2 × S 1 )# . . . #(S 2 × S 1 )#K1 # . . . #Kj

where the Si are spherical space forms (this uses the solution of the Poincare
conjecture) and the Ki are K(π, 1) manifolds (a K(π, 1) manifold is a closed
manifold with contractible universal cover and fundamental group isomorphic
to a group π .) Using the fill radius bound it can be shown that if, in addition, M
has positive scalar curvature then no K(π, 1) summands occur in this direct sum
decomposition (see [G-L2 ] for a proof). In particular, this implies that if M is a
closed three manifold with positive scalar curvature then the fundamental group
of M is virtually free. This verifies Conjecture 1 in the three dimensional case
since 2-positive Ricci curvature implies 3-positive Ricci curvature (positive
scalar curvature).
 Manifolds with k-positive Ricci curvature 197

The second result is due to Gromov-Lawson [G-L2]:

Theorem 9.3.2 Let (M, g) be a closed n-manifold with fill radius bounded
above by β. Then there is a distance decreasing map φ : M → " onto a metric
graph, such that, for each p ∈ ",

diameter(φ −1 (p)) ≤ C(β).

Proof The theorem follows from the proof of Corollary 9.11 in [G-L2]. Also
see [G] Appendix 1. 

The theorem implies that closed n-manifolds with fill radius bounded above
and large diameter are “long” and “thin”, exactly like compact 3-manifolds
with positive scalar curvature and large diameter.
We have already observed that in three dimensions 2-positive Ricci curvature
implies 3-positive Ricci curvature (positive scalar curvature). Therefore closed
three manifolds with 2-positive Ricci curvature bounded away from zero by α
satisfy a fill radius bound depending on α. This partly motivates the following
conjecture:

Conjecture 2 If M is a closed four manifold that admits a metric with 2-

positive Ricci curvature bounded below by α > 0 then M satisfies a fill radius
bound depending on α.

There is another “motivation” for this conjecture: Let (, ∂) be a stable
minimally immersed surface with boundary in a Riemannian four-manifold
(M, g). Using an averaging technique, the Gauss equation and an appropriate
choice of variational vector field, the second variation formula for area and
stability can be used to show that for any smooth function of compact support
f ∈ C0∞ () we have the inequalities:

1
|∇f |2 + f 2 K − Kν − (Ric11 + Ric22 ) eσ da ≥ 0 (9.3.12)
 2
and

1
|∇f |2 + f 2 K + Kν − (Ric11 + Ric22 ) e−σ da ≥ 0. (9.3.13)
 2
Here K is the Gauss curvature on  (in the induced metric), Kν is the curvature
of the normal bundle, Ric is the Ricci curvature on M, {e1 , e2 } is an orthonormal
frame on  and σ is a function on  that satisfies σ = Kν . Suppose that
(M, g) has 2-positive Ricci curvature bounded below by α > 0. Then,

Ric11 + Ric22 ≥ α.
198 J. Wolfson

In the case that the normal bundle is flat or, more generally, that the function σ
has suitably small oscillation, equations (9.3.12) and (9.3.13) imply:

1
|∇f |2 + f 2 K − (Ric11 + Ric22 ) da ≥ 0. (9.3.14)
 2

Then (9.3.14) can be used in the above argument from [S-Y2] to prove a
diameter bound on  in the induced metric. This implies a fill radius bound
and hence Conjecture 2. However, it can be shown that this line of reasoning
does not, in general, hold, the obstruction being the normal curvature. This
does not mean that a stable minimally immersed surface in a Riemannian
four-manifold (M, g) with 2-positive Ricci curvature bounded below does not
satisfy a diameter bound. Rather that the above reasoning does not apply. It
remains an interesting, if unexploited, fact that the 2-positive Ricci curvature
occurs in an averaged version of the second variation formula.
In the next section we will show that Conjecture 2 implies Conjecture 1 (in
the four dimensional case).

9.4 The fundamental group and fill radius bounds

The main theorem on the relation between fill radius and the fundamental group
can be stated:

Theorem 9.4.1 Let M be a closed Riemannian n-manifold. Suppose that the

universal cover π : M̃ → M is given the Riemannian metric g̃ such that π is
a local isometry. If (M̃, g̃) has bounded fill radius then the fundamental group
of M is virtually free.

Note, in particular, if a curvature condition implies a fill radius bound then

a closed Riemannian manifold, satisfying this curvature condition, satisfies the
hypotheses of the theorem. The proof of Theorem 9.4.1 is somewhat technical
and is done in [R-W]. In this survey we describe the main ideas and give a
complete proof under the additional hypothesis that π1 (M) is torsion free.
One of the main ideas in the proof of Theorem 9.4.1 concerns the number
of ends of a group.

Definition 9.4.2 Given a group G we define the number of ends, e(G), of G

to be the number of geometric ends of K̃, where K̃ → K is a regular covering
of the finite simplicial complex K by the simplicial complex K̃ and G is the
group of covering transformation.
 Manifolds with k-positive Ricci curvature 199

In particular, if G is the fundamental group of a closed manifold M then the

number of ends of G is the number of ends of the universal cover M̃ of M. It
is clear that a group G with no ends is virtually trivial and hence finite. It can
be shown [E] that a group with two ends is virtually infinite cyclic. A group
with three ends, in fact, has infinitely many ends. This can be seen by using the
Deck transformations on the universal cover. Thus a group G can have 0,1,2 or
infinitely many ends.
Our first result is on the fill radius and the number of ends of subgroups of
the fundamental group. We assume that π1 (M) is torsion free though the result
holds without this assumption [R-W]. We begin with a lemma.
Lemma 9.4.3 Let M be a closed manifold. Suppose that N → M is a covering
of M such that N has fundamental group G that is finitely generated and has
exactly one end. Let γ be a simple closed curve in N that represents an infinite
order generator [γ ] of G. Let M̃ → N be the universal cover and let γ̃ be the
lift of γ to M̃. Then the two ends of γ̃ lie in the same end of M̃.
Proof There is a finite simplicial complex K with regular covering K̃ such
that G acts as the group of covering transformations. There is an imbedding
ı : K → N that induces an epimorphism of fundamental groups. In particular,
the generators of G all lie in K. Then there is an imbedding ı̃ : K̃ → M̃. If
B ⊂ M̃ is compact then ı̃ −1 (B) ⊂ K̃ is compact.
Let γ be a simple closed curve in N that represents an infinite order generator
[γ ] of G. After a homotopy the lift γ̃ can be assumed to lie in K̃. Since G has
exactly one end, any two points on γ̃ , not in ı̃ −1 (B), can be joined by a curve
α in K̃ \ ı̃ −1 (B). The curve ı̃(α) then lies in M̃ \ B and joins points on γ̃ not
in B. Since this is true for any compact set B the conclusion follows. 
Proposition 9.4.4 Let M be a closed Riemannian manifold with torsion free
fundamental group. Suppose that the universal cover M̃of M has the property
that the fill radius of every null homotopic simple closed curve is uniformly
bounded above. Then no finitely generated subgroup of π1 (M) has exactly one
end.
Proof Assume, by way of contradiction, that a finitely generated subgroup G
of π1 (M) has exactly one end. Let N be a covering of M with fundamental
group π1 (N) isomorphic to G. Since π1 (M) is torsion free every element of G
is of infinite order. Let g ∈ G be a generator.
Denote by γ a closed minimal geodesic in N that represents g. Let p : M̃ →
N be the universal cover and let γ̃ be the geodesic line that is a lift to M̃ of
γ . Let x ∈ γ̃ and BR (x) ⊂ M̃ be the metric ball of radius R, center x. Then
because G has exactly one end by Lemma 9.4.3 both ends of γ̃ in M̃ \ BR (x)
200 J. Wolfson

lie in the same end of M̃. The geodesic line γ̃ consists of two geodesic rays
γ̃1 and γ̃2 beginning at x. For i = 1, 2, choose a point pi ∈ M̃ \ BR (x) along
γ̃i and denote the segment of γ̃i from x to pi by τi . Since p1 and p2 lie in the
same end there is a curve β ⊂ M̃ \ BR (x) joining p1 and p2 . Denote the closed
curve τ1 ∪ β ∪ τ2 by η. Since M̃ is simply connected η is null homotopic and
has fill radius greater than R2 . For sufficiently large R this contradicts the fill
radius bound. 

To illustrate the use of Proposition 9.4.4 we prove the following special case
of Theorem 9.4.1.

Theorem 9.4.5 Let M be a closed Riemannian manifold with torsion free

fundamental group. Suppose that the universal cover M̃of M has uniformly
bounded fill radius. Then π1 (M) is a free group of finite rank.
We will need the following theorem of Stallings [St]: If G is a torsion-free,
finitely generated group with infinitely many ends then G is a non-trivial free
product.

Proof Applying Stallings’ theorem to G = π1 (M), we have G  G1 ∗ G2 ,

where each Gi is finitely generated (by Grushko’s Theorem, see [Ma]). Each
Gi has either two or infinitely many ends (by Theorem 9.4.4). Then apply
Stallings theorem to each Gi with infinitely many ends and iterate. By Grushko’s
Theorem, this process terminates after finitely many steps resulting in G 
G1 ∗ · · · ∗ Gk , where each Gi is finitely generated and has two ends. Since
a torsion-free, finitely generated group with two ends is infinite cyclic, we
conclude that G = π1 (M) is a free group of finite rank. 

References
[E] Epstein, D., Ends, Topology of 3-manifolds edited by, M.K. Fort, Jr., Prentice-
Hall, 1962, 110–117.
[Ga] Gajer, J., Riemannian metrics of positive scalar curvature on compact manifolds
with boundary, Ann. Glob. Anal. Geom. 5 (1987) 179–191.
[G] Gromov, M, Filling Riemannian manifolds, J. Diff. Geom. 18 (1983) 1–147.
[G-L1] Gromov, M, and Lawson, H., The classification of simply connected manifolds
of positive scalar curvature, Ann of Math. 111 (1980) 423–434.
[G-L2] Gromov, M, and Lawson, H., Positive scalar curvature and the Dirac operator
on complete Riemannian manifolds, Publ. Math de IHES, 58 (1983) 83–196.
[Ma] Massey, W., Algebraic Topology: An Introduction, GTM 56, Springer-Verlag,
New York, 1984.
[R-W] Ramachandran, M. and Wolfson, J., Fill Radius and the Fundamental Group,
Journal of Topology and Analysis 2 (2010) 99–107.
 Manifolds with k-positive Ricci curvature 201

[R-S] Rosenberg, J., and Stolz, S., Metrics of positive scalar curvature and connections
with surgery, in Surveys on Surgery Theory, Volume 2, edited by S. Cappell,
et al., Annals of Math Studies 149, Princeton Univ. Press, Princeton, 2001.
[S-Y1] Schoen, R., and Yau, S. T., On the structure of manifolds of positive scalar
curvature, Manu. Math. 28 (1979) 159–183.
[S-Y2] Schoen, R., and Yau, S. T., The existence of a black hole due to condensation
of matter, Comm. Math. Phys. 90 (1983) 575–579.
[Sz] Stolz, S., Simply connected manifolds of positive scalar curvature, Ann. of
Math. 136 (1992), 511–540.
[St] Stallings, J., On torsion-free groups with infinitely many ends, Ann. of Math.
88 (1968), 312–334.

Author’s address:

Department of Mathematics,
Michigan State University,
East Lansing, MI 48824
USA
[email protected]